]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
e1b77c92 | 29 | #include <linux/kasan.h> |
1da177e4 LT |
30 | #include <linux/mm.h> |
31 | #include <linux/module.h> | |
32 | #include <linux/nmi.h> | |
33 | #include <linux/init.h> | |
dff06c15 | 34 | #include <linux/uaccess.h> |
1da177e4 | 35 | #include <linux/highmem.h> |
1da177e4 LT |
36 | #include <asm/mmu_context.h> |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 | 70 | #include <linux/ctype.h> |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
91d1aa43 | 74 | #include <linux/context_tracking.h> |
52f5684c | 75 | #include <linux/compiler.h> |
8e05e96a | 76 | #include <linux/frame.h> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
029632fb PZ |
93 | DEFINE_MUTEX(sched_domains_mutex); |
94 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 95 | |
fe44d621 | 96 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 97 | |
029632fb | 98 | void update_rq_clock(struct rq *rq) |
3e51f33f | 99 | { |
fe44d621 | 100 | s64 delta; |
305e6835 | 101 | |
9edfbfed PZ |
102 | lockdep_assert_held(&rq->lock); |
103 | ||
104 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 105 | return; |
aa483808 | 106 | |
fe44d621 | 107 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
108 | if (delta < 0) |
109 | return; | |
fe44d621 PZ |
110 | rq->clock += delta; |
111 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
112 | } |
113 | ||
bf5c91ba IM |
114 | /* |
115 | * Debugging: various feature bits | |
116 | */ | |
f00b45c1 | 117 | |
f00b45c1 PZ |
118 | #define SCHED_FEAT(name, enabled) \ |
119 | (1UL << __SCHED_FEAT_##name) * enabled | | |
120 | ||
bf5c91ba | 121 | const_debug unsigned int sysctl_sched_features = |
391e43da | 122 | #include "features.h" |
f00b45c1 PZ |
123 | 0; |
124 | ||
125 | #undef SCHED_FEAT | |
126 | ||
b82d9fdd PZ |
127 | /* |
128 | * Number of tasks to iterate in a single balance run. | |
129 | * Limited because this is done with IRQs disabled. | |
130 | */ | |
131 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
132 | ||
e9e9250b PZ |
133 | /* |
134 | * period over which we average the RT time consumption, measured | |
135 | * in ms. | |
136 | * | |
137 | * default: 1s | |
138 | */ | |
139 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
140 | ||
fa85ae24 | 141 | /* |
9f0c1e56 | 142 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
143 | * default: 1s |
144 | */ | |
9f0c1e56 | 145 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 146 | |
029632fb | 147 | __read_mostly int scheduler_running; |
6892b75e | 148 | |
9f0c1e56 PZ |
149 | /* |
150 | * part of the period that we allow rt tasks to run in us. | |
151 | * default: 0.95s | |
152 | */ | |
153 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 154 | |
3fa0818b RR |
155 | /* cpus with isolated domains */ |
156 | cpumask_var_t cpu_isolated_map; | |
157 | ||
1da177e4 | 158 | /* |
cc2a73b5 | 159 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 160 | */ |
a9957449 | 161 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
162 | __acquires(rq->lock) |
163 | { | |
70b97a7f | 164 | struct rq *rq; |
1da177e4 LT |
165 | |
166 | local_irq_disable(); | |
167 | rq = this_rq(); | |
05fa785c | 168 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
169 | |
170 | return rq; | |
171 | } | |
172 | ||
8f4d37ec PZ |
173 | #ifdef CONFIG_SCHED_HRTICK |
174 | /* | |
175 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 176 | */ |
8f4d37ec | 177 | |
8f4d37ec PZ |
178 | static void hrtick_clear(struct rq *rq) |
179 | { | |
180 | if (hrtimer_active(&rq->hrtick_timer)) | |
181 | hrtimer_cancel(&rq->hrtick_timer); | |
182 | } | |
183 | ||
8f4d37ec PZ |
184 | /* |
185 | * High-resolution timer tick. | |
186 | * Runs from hardirq context with interrupts disabled. | |
187 | */ | |
188 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
189 | { | |
190 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
191 | ||
192 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
193 | ||
05fa785c | 194 | raw_spin_lock(&rq->lock); |
3e51f33f | 195 | update_rq_clock(rq); |
8f4d37ec | 196 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 197 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
198 | |
199 | return HRTIMER_NORESTART; | |
200 | } | |
201 | ||
95e904c7 | 202 | #ifdef CONFIG_SMP |
971ee28c | 203 | |
4961b6e1 | 204 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
205 | { |
206 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 207 | |
4961b6e1 | 208 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
209 | } |
210 | ||
31656519 PZ |
211 | /* |
212 | * called from hardirq (IPI) context | |
213 | */ | |
214 | static void __hrtick_start(void *arg) | |
b328ca18 | 215 | { |
31656519 | 216 | struct rq *rq = arg; |
b328ca18 | 217 | |
05fa785c | 218 | raw_spin_lock(&rq->lock); |
971ee28c | 219 | __hrtick_restart(rq); |
31656519 | 220 | rq->hrtick_csd_pending = 0; |
05fa785c | 221 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
222 | } |
223 | ||
31656519 PZ |
224 | /* |
225 | * Called to set the hrtick timer state. | |
226 | * | |
227 | * called with rq->lock held and irqs disabled | |
228 | */ | |
029632fb | 229 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 230 | { |
31656519 | 231 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 232 | ktime_t time; |
233 | s64 delta; | |
234 | ||
235 | /* | |
236 | * Don't schedule slices shorter than 10000ns, that just | |
237 | * doesn't make sense and can cause timer DoS. | |
238 | */ | |
239 | delta = max_t(s64, delay, 10000LL); | |
240 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 241 | |
cc584b21 | 242 | hrtimer_set_expires(timer, time); |
31656519 PZ |
243 | |
244 | if (rq == this_rq()) { | |
971ee28c | 245 | __hrtick_restart(rq); |
31656519 | 246 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 247 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
248 | rq->hrtick_csd_pending = 1; |
249 | } | |
b328ca18 PZ |
250 | } |
251 | ||
252 | static int | |
253 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
254 | { | |
255 | int cpu = (int)(long)hcpu; | |
256 | ||
257 | switch (action) { | |
258 | case CPU_UP_CANCELED: | |
259 | case CPU_UP_CANCELED_FROZEN: | |
260 | case CPU_DOWN_PREPARE: | |
261 | case CPU_DOWN_PREPARE_FROZEN: | |
262 | case CPU_DEAD: | |
263 | case CPU_DEAD_FROZEN: | |
31656519 | 264 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
265 | return NOTIFY_OK; |
266 | } | |
267 | ||
268 | return NOTIFY_DONE; | |
269 | } | |
270 | ||
fa748203 | 271 | static __init void init_hrtick(void) |
b328ca18 PZ |
272 | { |
273 | hotcpu_notifier(hotplug_hrtick, 0); | |
274 | } | |
31656519 PZ |
275 | #else |
276 | /* | |
277 | * Called to set the hrtick timer state. | |
278 | * | |
279 | * called with rq->lock held and irqs disabled | |
280 | */ | |
029632fb | 281 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 282 | { |
86893335 WL |
283 | /* |
284 | * Don't schedule slices shorter than 10000ns, that just | |
285 | * doesn't make sense. Rely on vruntime for fairness. | |
286 | */ | |
287 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
288 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
289 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 290 | } |
b328ca18 | 291 | |
006c75f1 | 292 | static inline void init_hrtick(void) |
8f4d37ec | 293 | { |
8f4d37ec | 294 | } |
31656519 | 295 | #endif /* CONFIG_SMP */ |
8f4d37ec | 296 | |
31656519 | 297 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 298 | { |
31656519 PZ |
299 | #ifdef CONFIG_SMP |
300 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 301 | |
31656519 PZ |
302 | rq->hrtick_csd.flags = 0; |
303 | rq->hrtick_csd.func = __hrtick_start; | |
304 | rq->hrtick_csd.info = rq; | |
305 | #endif | |
8f4d37ec | 306 | |
31656519 PZ |
307 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
308 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 309 | } |
006c75f1 | 310 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
311 | static inline void hrtick_clear(struct rq *rq) |
312 | { | |
313 | } | |
314 | ||
8f4d37ec PZ |
315 | static inline void init_rq_hrtick(struct rq *rq) |
316 | { | |
317 | } | |
318 | ||
b328ca18 PZ |
319 | static inline void init_hrtick(void) |
320 | { | |
321 | } | |
006c75f1 | 322 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 323 | |
5529578a FW |
324 | /* |
325 | * cmpxchg based fetch_or, macro so it works for different integer types | |
326 | */ | |
327 | #define fetch_or(ptr, mask) \ | |
328 | ({ \ | |
329 | typeof(ptr) _ptr = (ptr); \ | |
330 | typeof(mask) _mask = (mask); \ | |
331 | typeof(*_ptr) _old, _val = *_ptr; \ | |
332 | \ | |
333 | for (;;) { \ | |
334 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
335 | if (_old == _val) \ | |
336 | break; \ | |
337 | _val = _old; \ | |
338 | } \ | |
339 | _old; \ | |
340 | }) | |
341 | ||
e3baac47 | 342 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
343 | /* |
344 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
345 | * this avoids any races wrt polling state changes and thereby avoids | |
346 | * spurious IPIs. | |
347 | */ | |
348 | static bool set_nr_and_not_polling(struct task_struct *p) | |
349 | { | |
350 | struct thread_info *ti = task_thread_info(p); | |
351 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
352 | } | |
e3baac47 PZ |
353 | |
354 | /* | |
355 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
356 | * | |
357 | * If this returns true, then the idle task promises to call | |
358 | * sched_ttwu_pending() and reschedule soon. | |
359 | */ | |
360 | static bool set_nr_if_polling(struct task_struct *p) | |
361 | { | |
362 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 363 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
364 | |
365 | for (;;) { | |
366 | if (!(val & _TIF_POLLING_NRFLAG)) | |
367 | return false; | |
368 | if (val & _TIF_NEED_RESCHED) | |
369 | return true; | |
370 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
371 | if (old == val) | |
372 | break; | |
373 | val = old; | |
374 | } | |
375 | return true; | |
376 | } | |
377 | ||
fd99f91a PZ |
378 | #else |
379 | static bool set_nr_and_not_polling(struct task_struct *p) | |
380 | { | |
381 | set_tsk_need_resched(p); | |
382 | return true; | |
383 | } | |
e3baac47 PZ |
384 | |
385 | #ifdef CONFIG_SMP | |
386 | static bool set_nr_if_polling(struct task_struct *p) | |
387 | { | |
388 | return false; | |
389 | } | |
390 | #endif | |
fd99f91a PZ |
391 | #endif |
392 | ||
76751049 PZ |
393 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
394 | { | |
395 | struct wake_q_node *node = &task->wake_q; | |
396 | ||
397 | /* | |
398 | * Atomically grab the task, if ->wake_q is !nil already it means | |
399 | * its already queued (either by us or someone else) and will get the | |
400 | * wakeup due to that. | |
401 | * | |
402 | * This cmpxchg() implies a full barrier, which pairs with the write | |
403 | * barrier implied by the wakeup in wake_up_list(). | |
404 | */ | |
405 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
406 | return; | |
407 | ||
408 | get_task_struct(task); | |
409 | ||
410 | /* | |
411 | * The head is context local, there can be no concurrency. | |
412 | */ | |
413 | *head->lastp = node; | |
414 | head->lastp = &node->next; | |
415 | } | |
416 | ||
417 | void wake_up_q(struct wake_q_head *head) | |
418 | { | |
419 | struct wake_q_node *node = head->first; | |
420 | ||
421 | while (node != WAKE_Q_TAIL) { | |
422 | struct task_struct *task; | |
423 | ||
424 | task = container_of(node, struct task_struct, wake_q); | |
425 | BUG_ON(!task); | |
426 | /* task can safely be re-inserted now */ | |
427 | node = node->next; | |
428 | task->wake_q.next = NULL; | |
429 | ||
430 | /* | |
431 | * wake_up_process() implies a wmb() to pair with the queueing | |
432 | * in wake_q_add() so as not to miss wakeups. | |
433 | */ | |
434 | wake_up_process(task); | |
435 | put_task_struct(task); | |
436 | } | |
437 | } | |
438 | ||
c24d20db | 439 | /* |
8875125e | 440 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
441 | * |
442 | * On UP this means the setting of the need_resched flag, on SMP it | |
443 | * might also involve a cross-CPU call to trigger the scheduler on | |
444 | * the target CPU. | |
445 | */ | |
8875125e | 446 | void resched_curr(struct rq *rq) |
c24d20db | 447 | { |
8875125e | 448 | struct task_struct *curr = rq->curr; |
c24d20db IM |
449 | int cpu; |
450 | ||
8875125e | 451 | lockdep_assert_held(&rq->lock); |
c24d20db | 452 | |
8875125e | 453 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
454 | return; |
455 | ||
8875125e | 456 | cpu = cpu_of(rq); |
fd99f91a | 457 | |
f27dde8d | 458 | if (cpu == smp_processor_id()) { |
8875125e | 459 | set_tsk_need_resched(curr); |
f27dde8d | 460 | set_preempt_need_resched(); |
c24d20db | 461 | return; |
f27dde8d | 462 | } |
c24d20db | 463 | |
8875125e | 464 | if (set_nr_and_not_polling(curr)) |
c24d20db | 465 | smp_send_reschedule(cpu); |
dfc68f29 AL |
466 | else |
467 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
468 | } |
469 | ||
029632fb | 470 | void resched_cpu(int cpu) |
c24d20db IM |
471 | { |
472 | struct rq *rq = cpu_rq(cpu); | |
473 | unsigned long flags; | |
474 | ||
05fa785c | 475 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 476 | return; |
8875125e | 477 | resched_curr(rq); |
05fa785c | 478 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 479 | } |
06d8308c | 480 | |
b021fe3e | 481 | #ifdef CONFIG_SMP |
3451d024 | 482 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
483 | /* |
484 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
485 | * from an idle cpu. This is good for power-savings. | |
486 | * | |
487 | * We don't do similar optimization for completely idle system, as | |
488 | * selecting an idle cpu will add more delays to the timers than intended | |
489 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
490 | */ | |
bc7a34b8 | 491 | int get_nohz_timer_target(void) |
83cd4fe2 | 492 | { |
bc7a34b8 | 493 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
494 | struct sched_domain *sd; |
495 | ||
9642d18e | 496 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
497 | return cpu; |
498 | ||
057f3fad | 499 | rcu_read_lock(); |
83cd4fe2 | 500 | for_each_domain(cpu, sd) { |
057f3fad | 501 | for_each_cpu(i, sched_domain_span(sd)) { |
9642d18e | 502 | if (!idle_cpu(i) && is_housekeeping_cpu(cpu)) { |
057f3fad PZ |
503 | cpu = i; |
504 | goto unlock; | |
505 | } | |
506 | } | |
83cd4fe2 | 507 | } |
9642d18e VH |
508 | |
509 | if (!is_housekeeping_cpu(cpu)) | |
510 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
511 | unlock: |
512 | rcu_read_unlock(); | |
83cd4fe2 VP |
513 | return cpu; |
514 | } | |
06d8308c TG |
515 | /* |
516 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
517 | * idle CPU then this timer might expire before the next timer event | |
518 | * which is scheduled to wake up that CPU. In case of a completely | |
519 | * idle system the next event might even be infinite time into the | |
520 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
521 | * leaves the inner idle loop so the newly added timer is taken into | |
522 | * account when the CPU goes back to idle and evaluates the timer | |
523 | * wheel for the next timer event. | |
524 | */ | |
1c20091e | 525 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
526 | { |
527 | struct rq *rq = cpu_rq(cpu); | |
528 | ||
529 | if (cpu == smp_processor_id()) | |
530 | return; | |
531 | ||
67b9ca70 | 532 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 533 | smp_send_reschedule(cpu); |
dfc68f29 AL |
534 | else |
535 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
536 | } |
537 | ||
c5bfece2 | 538 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 539 | { |
53c5fa16 FW |
540 | /* |
541 | * We just need the target to call irq_exit() and re-evaluate | |
542 | * the next tick. The nohz full kick at least implies that. | |
543 | * If needed we can still optimize that later with an | |
544 | * empty IRQ. | |
545 | */ | |
c5bfece2 | 546 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
547 | if (cpu != smp_processor_id() || |
548 | tick_nohz_tick_stopped()) | |
53c5fa16 | 549 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
550 | return true; |
551 | } | |
552 | ||
553 | return false; | |
554 | } | |
555 | ||
556 | void wake_up_nohz_cpu(int cpu) | |
557 | { | |
c5bfece2 | 558 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
559 | wake_up_idle_cpu(cpu); |
560 | } | |
561 | ||
ca38062e | 562 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 563 | { |
1c792db7 | 564 | int cpu = smp_processor_id(); |
873b4c65 VG |
565 | |
566 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
567 | return false; | |
568 | ||
569 | if (idle_cpu(cpu) && !need_resched()) | |
570 | return true; | |
571 | ||
572 | /* | |
573 | * We can't run Idle Load Balance on this CPU for this time so we | |
574 | * cancel it and clear NOHZ_BALANCE_KICK | |
575 | */ | |
576 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
577 | return false; | |
45bf76df IM |
578 | } |
579 | ||
3451d024 | 580 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 581 | |
ca38062e | 582 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 583 | { |
ca38062e | 584 | return false; |
2069dd75 PZ |
585 | } |
586 | ||
3451d024 | 587 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 588 | |
ce831b38 | 589 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 590 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 591 | { |
76d92ac3 FW |
592 | int fifo_nr_running; |
593 | ||
594 | /* Deadline tasks, even if single, need the tick */ | |
595 | if (rq->dl.dl_nr_running) | |
596 | return false; | |
597 | ||
1e78cdbd | 598 | /* |
76d92ac3 FW |
599 | * FIFO realtime policy runs the highest priority task (after DEADLINE). |
600 | * Other runnable tasks are of a lower priority. The scheduler tick | |
601 | * isn't needed. | |
1e78cdbd | 602 | */ |
76d92ac3 FW |
603 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; |
604 | if (fifo_nr_running) | |
1e78cdbd RR |
605 | return true; |
606 | ||
607 | /* | |
608 | * Round-robin realtime tasks time slice with other tasks at the same | |
76d92ac3 | 609 | * realtime priority. |
1e78cdbd | 610 | */ |
76d92ac3 FW |
611 | if (rq->rt.rr_nr_running) { |
612 | if (rq->rt.rr_nr_running == 1) | |
613 | return true; | |
614 | else | |
615 | return false; | |
1e78cdbd RR |
616 | } |
617 | ||
76d92ac3 FW |
618 | /* Normal multitasking need periodic preemption checks */ |
619 | if (rq->cfs.nr_running > 1) | |
541b8264 | 620 | return false; |
ce831b38 | 621 | |
541b8264 | 622 | return true; |
ce831b38 FW |
623 | } |
624 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 625 | |
029632fb | 626 | void sched_avg_update(struct rq *rq) |
18d95a28 | 627 | { |
e9e9250b PZ |
628 | s64 period = sched_avg_period(); |
629 | ||
78becc27 | 630 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
631 | /* |
632 | * Inline assembly required to prevent the compiler | |
633 | * optimising this loop into a divmod call. | |
634 | * See __iter_div_u64_rem() for another example of this. | |
635 | */ | |
636 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
637 | rq->age_stamp += period; |
638 | rq->rt_avg /= 2; | |
639 | } | |
18d95a28 PZ |
640 | } |
641 | ||
6d6bc0ad | 642 | #endif /* CONFIG_SMP */ |
18d95a28 | 643 | |
a790de99 PT |
644 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
645 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 646 | /* |
8277434e PT |
647 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
648 | * node and @up when leaving it for the final time. | |
649 | * | |
650 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 651 | */ |
029632fb | 652 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 653 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
654 | { |
655 | struct task_group *parent, *child; | |
eb755805 | 656 | int ret; |
c09595f6 | 657 | |
8277434e PT |
658 | parent = from; |
659 | ||
c09595f6 | 660 | down: |
eb755805 PZ |
661 | ret = (*down)(parent, data); |
662 | if (ret) | |
8277434e | 663 | goto out; |
c09595f6 PZ |
664 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
665 | parent = child; | |
666 | goto down; | |
667 | ||
668 | up: | |
669 | continue; | |
670 | } | |
eb755805 | 671 | ret = (*up)(parent, data); |
8277434e PT |
672 | if (ret || parent == from) |
673 | goto out; | |
c09595f6 PZ |
674 | |
675 | child = parent; | |
676 | parent = parent->parent; | |
677 | if (parent) | |
678 | goto up; | |
8277434e | 679 | out: |
eb755805 | 680 | return ret; |
c09595f6 PZ |
681 | } |
682 | ||
029632fb | 683 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 684 | { |
e2b245f8 | 685 | return 0; |
eb755805 | 686 | } |
18d95a28 PZ |
687 | #endif |
688 | ||
45bf76df IM |
689 | static void set_load_weight(struct task_struct *p) |
690 | { | |
f05998d4 NR |
691 | int prio = p->static_prio - MAX_RT_PRIO; |
692 | struct load_weight *load = &p->se.load; | |
693 | ||
dd41f596 IM |
694 | /* |
695 | * SCHED_IDLE tasks get minimal weight: | |
696 | */ | |
20f9cd2a | 697 | if (idle_policy(p->policy)) { |
c8b28116 | 698 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 699 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
700 | return; |
701 | } | |
71f8bd46 | 702 | |
ed82b8a1 AK |
703 | load->weight = scale_load(sched_prio_to_weight[prio]); |
704 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
705 | } |
706 | ||
1de64443 | 707 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 708 | { |
a64692a3 | 709 | update_rq_clock(rq); |
1de64443 PZ |
710 | if (!(flags & ENQUEUE_RESTORE)) |
711 | sched_info_queued(rq, p); | |
371fd7e7 | 712 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
713 | } |
714 | ||
1de64443 | 715 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 716 | { |
a64692a3 | 717 | update_rq_clock(rq); |
1de64443 PZ |
718 | if (!(flags & DEQUEUE_SAVE)) |
719 | sched_info_dequeued(rq, p); | |
371fd7e7 | 720 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
721 | } |
722 | ||
029632fb | 723 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
724 | { |
725 | if (task_contributes_to_load(p)) | |
726 | rq->nr_uninterruptible--; | |
727 | ||
371fd7e7 | 728 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
729 | } |
730 | ||
029632fb | 731 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
732 | { |
733 | if (task_contributes_to_load(p)) | |
734 | rq->nr_uninterruptible++; | |
735 | ||
371fd7e7 | 736 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
737 | } |
738 | ||
fe44d621 | 739 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 740 | { |
095c0aa8 GC |
741 | /* |
742 | * In theory, the compile should just see 0 here, and optimize out the call | |
743 | * to sched_rt_avg_update. But I don't trust it... | |
744 | */ | |
745 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
746 | s64 steal = 0, irq_delta = 0; | |
747 | #endif | |
748 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 749 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
750 | |
751 | /* | |
752 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
753 | * this case when a previous update_rq_clock() happened inside a | |
754 | * {soft,}irq region. | |
755 | * | |
756 | * When this happens, we stop ->clock_task and only update the | |
757 | * prev_irq_time stamp to account for the part that fit, so that a next | |
758 | * update will consume the rest. This ensures ->clock_task is | |
759 | * monotonic. | |
760 | * | |
761 | * It does however cause some slight miss-attribution of {soft,}irq | |
762 | * time, a more accurate solution would be to update the irq_time using | |
763 | * the current rq->clock timestamp, except that would require using | |
764 | * atomic ops. | |
765 | */ | |
766 | if (irq_delta > delta) | |
767 | irq_delta = delta; | |
768 | ||
769 | rq->prev_irq_time += irq_delta; | |
770 | delta -= irq_delta; | |
095c0aa8 GC |
771 | #endif |
772 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 773 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
774 | steal = paravirt_steal_clock(cpu_of(rq)); |
775 | steal -= rq->prev_steal_time_rq; | |
776 | ||
777 | if (unlikely(steal > delta)) | |
778 | steal = delta; | |
779 | ||
095c0aa8 | 780 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
781 | delta -= steal; |
782 | } | |
783 | #endif | |
784 | ||
fe44d621 PZ |
785 | rq->clock_task += delta; |
786 | ||
095c0aa8 | 787 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 788 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
789 | sched_rt_avg_update(rq, irq_delta + steal); |
790 | #endif | |
aa483808 VP |
791 | } |
792 | ||
34f971f6 PZ |
793 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
794 | { | |
795 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
796 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
797 | ||
798 | if (stop) { | |
799 | /* | |
800 | * Make it appear like a SCHED_FIFO task, its something | |
801 | * userspace knows about and won't get confused about. | |
802 | * | |
803 | * Also, it will make PI more or less work without too | |
804 | * much confusion -- but then, stop work should not | |
805 | * rely on PI working anyway. | |
806 | */ | |
807 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
808 | ||
809 | stop->sched_class = &stop_sched_class; | |
810 | } | |
811 | ||
812 | cpu_rq(cpu)->stop = stop; | |
813 | ||
814 | if (old_stop) { | |
815 | /* | |
816 | * Reset it back to a normal scheduling class so that | |
817 | * it can die in pieces. | |
818 | */ | |
819 | old_stop->sched_class = &rt_sched_class; | |
820 | } | |
821 | } | |
822 | ||
14531189 | 823 | /* |
dd41f596 | 824 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 825 | */ |
14531189 IM |
826 | static inline int __normal_prio(struct task_struct *p) |
827 | { | |
dd41f596 | 828 | return p->static_prio; |
14531189 IM |
829 | } |
830 | ||
b29739f9 IM |
831 | /* |
832 | * Calculate the expected normal priority: i.e. priority | |
833 | * without taking RT-inheritance into account. Might be | |
834 | * boosted by interactivity modifiers. Changes upon fork, | |
835 | * setprio syscalls, and whenever the interactivity | |
836 | * estimator recalculates. | |
837 | */ | |
36c8b586 | 838 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
839 | { |
840 | int prio; | |
841 | ||
aab03e05 DF |
842 | if (task_has_dl_policy(p)) |
843 | prio = MAX_DL_PRIO-1; | |
844 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
845 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
846 | else | |
847 | prio = __normal_prio(p); | |
848 | return prio; | |
849 | } | |
850 | ||
851 | /* | |
852 | * Calculate the current priority, i.e. the priority | |
853 | * taken into account by the scheduler. This value might | |
854 | * be boosted by RT tasks, or might be boosted by | |
855 | * interactivity modifiers. Will be RT if the task got | |
856 | * RT-boosted. If not then it returns p->normal_prio. | |
857 | */ | |
36c8b586 | 858 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
859 | { |
860 | p->normal_prio = normal_prio(p); | |
861 | /* | |
862 | * If we are RT tasks or we were boosted to RT priority, | |
863 | * keep the priority unchanged. Otherwise, update priority | |
864 | * to the normal priority: | |
865 | */ | |
866 | if (!rt_prio(p->prio)) | |
867 | return p->normal_prio; | |
868 | return p->prio; | |
869 | } | |
870 | ||
1da177e4 LT |
871 | /** |
872 | * task_curr - is this task currently executing on a CPU? | |
873 | * @p: the task in question. | |
e69f6186 YB |
874 | * |
875 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 876 | */ |
36c8b586 | 877 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
878 | { |
879 | return cpu_curr(task_cpu(p)) == p; | |
880 | } | |
881 | ||
67dfa1b7 | 882 | /* |
4c9a4bc8 PZ |
883 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
884 | * use the balance_callback list if you want balancing. | |
885 | * | |
886 | * this means any call to check_class_changed() must be followed by a call to | |
887 | * balance_callback(). | |
67dfa1b7 | 888 | */ |
cb469845 SR |
889 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
890 | const struct sched_class *prev_class, | |
da7a735e | 891 | int oldprio) |
cb469845 SR |
892 | { |
893 | if (prev_class != p->sched_class) { | |
894 | if (prev_class->switched_from) | |
da7a735e | 895 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 896 | |
da7a735e | 897 | p->sched_class->switched_to(rq, p); |
2d3d891d | 898 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 899 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
900 | } |
901 | ||
029632fb | 902 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
903 | { |
904 | const struct sched_class *class; | |
905 | ||
906 | if (p->sched_class == rq->curr->sched_class) { | |
907 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
908 | } else { | |
909 | for_each_class(class) { | |
910 | if (class == rq->curr->sched_class) | |
911 | break; | |
912 | if (class == p->sched_class) { | |
8875125e | 913 | resched_curr(rq); |
1e5a7405 PZ |
914 | break; |
915 | } | |
916 | } | |
917 | } | |
918 | ||
919 | /* | |
920 | * A queue event has occurred, and we're going to schedule. In | |
921 | * this case, we can save a useless back to back clock update. | |
922 | */ | |
da0c1e65 | 923 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 924 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
925 | } |
926 | ||
1da177e4 | 927 | #ifdef CONFIG_SMP |
5cc389bc PZ |
928 | /* |
929 | * This is how migration works: | |
930 | * | |
931 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
932 | * stop_one_cpu(). | |
933 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
934 | * off the CPU) | |
935 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
936 | * 4) if it's in the wrong runqueue then the migration thread removes | |
937 | * it and puts it into the right queue. | |
938 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
939 | * is done. | |
940 | */ | |
941 | ||
942 | /* | |
943 | * move_queued_task - move a queued task to new rq. | |
944 | * | |
945 | * Returns (locked) new rq. Old rq's lock is released. | |
946 | */ | |
5e16bbc2 | 947 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 948 | { |
5cc389bc PZ |
949 | lockdep_assert_held(&rq->lock); |
950 | ||
5cc389bc | 951 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 952 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
953 | set_task_cpu(p, new_cpu); |
954 | raw_spin_unlock(&rq->lock); | |
955 | ||
956 | rq = cpu_rq(new_cpu); | |
957 | ||
958 | raw_spin_lock(&rq->lock); | |
959 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 960 | enqueue_task(rq, p, 0); |
3ea94de1 | 961 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
962 | check_preempt_curr(rq, p, 0); |
963 | ||
964 | return rq; | |
965 | } | |
966 | ||
967 | struct migration_arg { | |
968 | struct task_struct *task; | |
969 | int dest_cpu; | |
970 | }; | |
971 | ||
972 | /* | |
973 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
974 | * this because either it can't run here any more (set_cpus_allowed() | |
975 | * away from this CPU, or CPU going down), or because we're | |
976 | * attempting to rebalance this task on exec (sched_exec). | |
977 | * | |
978 | * So we race with normal scheduler movements, but that's OK, as long | |
979 | * as the task is no longer on this CPU. | |
5cc389bc | 980 | */ |
5e16bbc2 | 981 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 982 | { |
5cc389bc | 983 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 984 | return rq; |
5cc389bc PZ |
985 | |
986 | /* Affinity changed (again). */ | |
987 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 988 | return rq; |
5cc389bc | 989 | |
5e16bbc2 PZ |
990 | rq = move_queued_task(rq, p, dest_cpu); |
991 | ||
992 | return rq; | |
5cc389bc PZ |
993 | } |
994 | ||
995 | /* | |
996 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
997 | * and performs thread migration by bumping thread off CPU then | |
998 | * 'pushing' onto another runqueue. | |
999 | */ | |
1000 | static int migration_cpu_stop(void *data) | |
1001 | { | |
1002 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1003 | struct task_struct *p = arg->task; |
1004 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1005 | |
1006 | /* | |
1007 | * The original target cpu might have gone down and we might | |
1008 | * be on another cpu but it doesn't matter. | |
1009 | */ | |
1010 | local_irq_disable(); | |
1011 | /* | |
1012 | * We need to explicitly wake pending tasks before running | |
1013 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1014 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1015 | */ | |
1016 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1017 | |
1018 | raw_spin_lock(&p->pi_lock); | |
1019 | raw_spin_lock(&rq->lock); | |
1020 | /* | |
1021 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1022 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1023 | * we're holding p->pi_lock. | |
1024 | */ | |
1025 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1026 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1027 | raw_spin_unlock(&rq->lock); | |
1028 | raw_spin_unlock(&p->pi_lock); | |
1029 | ||
5cc389bc PZ |
1030 | local_irq_enable(); |
1031 | return 0; | |
1032 | } | |
1033 | ||
c5b28038 PZ |
1034 | /* |
1035 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1036 | * actually call this function. | |
1037 | */ | |
1038 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1039 | { |
5cc389bc PZ |
1040 | cpumask_copy(&p->cpus_allowed, new_mask); |
1041 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1042 | } | |
1043 | ||
c5b28038 PZ |
1044 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1045 | { | |
6c37067e PZ |
1046 | struct rq *rq = task_rq(p); |
1047 | bool queued, running; | |
1048 | ||
c5b28038 | 1049 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1050 | |
1051 | queued = task_on_rq_queued(p); | |
1052 | running = task_current(rq, p); | |
1053 | ||
1054 | if (queued) { | |
1055 | /* | |
1056 | * Because __kthread_bind() calls this on blocked tasks without | |
1057 | * holding rq->lock. | |
1058 | */ | |
1059 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1060 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1061 | } |
1062 | if (running) | |
1063 | put_prev_task(rq, p); | |
1064 | ||
c5b28038 | 1065 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e PZ |
1066 | |
1067 | if (running) | |
1068 | p->sched_class->set_curr_task(rq); | |
1069 | if (queued) | |
1de64443 | 1070 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
c5b28038 PZ |
1071 | } |
1072 | ||
5cc389bc PZ |
1073 | /* |
1074 | * Change a given task's CPU affinity. Migrate the thread to a | |
1075 | * proper CPU and schedule it away if the CPU it's executing on | |
1076 | * is removed from the allowed bitmask. | |
1077 | * | |
1078 | * NOTE: the caller must have a valid reference to the task, the | |
1079 | * task must not exit() & deallocate itself prematurely. The | |
1080 | * call is not atomic; no spinlocks may be held. | |
1081 | */ | |
25834c73 PZ |
1082 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1083 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1084 | { |
e9d867a6 PZI |
1085 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
1086 | unsigned int dest_cpu; | |
5cc389bc PZ |
1087 | unsigned long flags; |
1088 | struct rq *rq; | |
5cc389bc PZ |
1089 | int ret = 0; |
1090 | ||
1091 | rq = task_rq_lock(p, &flags); | |
1092 | ||
e9d867a6 PZI |
1093 | if (p->flags & PF_KTHREAD) { |
1094 | /* | |
1095 | * Kernel threads are allowed on online && !active CPUs | |
1096 | */ | |
1097 | cpu_valid_mask = cpu_online_mask; | |
1098 | } | |
1099 | ||
25834c73 PZ |
1100 | /* |
1101 | * Must re-check here, to close a race against __kthread_bind(), | |
1102 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1103 | */ | |
1104 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1105 | ret = -EINVAL; | |
1106 | goto out; | |
1107 | } | |
1108 | ||
5cc389bc PZ |
1109 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1110 | goto out; | |
1111 | ||
e9d867a6 | 1112 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1113 | ret = -EINVAL; |
1114 | goto out; | |
1115 | } | |
1116 | ||
1117 | do_set_cpus_allowed(p, new_mask); | |
1118 | ||
e9d867a6 PZI |
1119 | if (p->flags & PF_KTHREAD) { |
1120 | /* | |
1121 | * For kernel threads that do indeed end up on online && | |
1122 | * !active we want to ensure they are strict per-cpu threads. | |
1123 | */ | |
1124 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1125 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1126 | p->nr_cpus_allowed != 1); | |
1127 | } | |
1128 | ||
5cc389bc PZ |
1129 | /* Can the task run on the task's current CPU? If so, we're done */ |
1130 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1131 | goto out; | |
1132 | ||
e9d867a6 | 1133 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1134 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1135 | struct migration_arg arg = { p, dest_cpu }; | |
1136 | /* Need help from migration thread: drop lock and wait. */ | |
1137 | task_rq_unlock(rq, p, &flags); | |
1138 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
1139 | tlb_migrate_finish(p->mm); | |
1140 | return 0; | |
cbce1a68 PZ |
1141 | } else if (task_on_rq_queued(p)) { |
1142 | /* | |
1143 | * OK, since we're going to drop the lock immediately | |
1144 | * afterwards anyway. | |
1145 | */ | |
1146 | lockdep_unpin_lock(&rq->lock); | |
5e16bbc2 | 1147 | rq = move_queued_task(rq, p, dest_cpu); |
cbce1a68 PZ |
1148 | lockdep_pin_lock(&rq->lock); |
1149 | } | |
5cc389bc PZ |
1150 | out: |
1151 | task_rq_unlock(rq, p, &flags); | |
1152 | ||
1153 | return ret; | |
1154 | } | |
25834c73 PZ |
1155 | |
1156 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1157 | { | |
1158 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1159 | } | |
5cc389bc PZ |
1160 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1161 | ||
dd41f596 | 1162 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1163 | { |
e2912009 PZ |
1164 | #ifdef CONFIG_SCHED_DEBUG |
1165 | /* | |
1166 | * We should never call set_task_cpu() on a blocked task, | |
1167 | * ttwu() will sort out the placement. | |
1168 | */ | |
077614ee | 1169 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1170 | !p->on_rq); |
0122ec5b | 1171 | |
3ea94de1 JP |
1172 | /* |
1173 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1174 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1175 | * time relying on p->on_rq. | |
1176 | */ | |
1177 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1178 | p->sched_class == &fair_sched_class && | |
1179 | (p->on_rq && !task_on_rq_migrating(p))); | |
1180 | ||
0122ec5b | 1181 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1182 | /* |
1183 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1184 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1185 | * | |
1186 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1187 | * see task_group(). |
6c6c54e1 PZ |
1188 | * |
1189 | * Furthermore, all task_rq users should acquire both locks, see | |
1190 | * task_rq_lock(). | |
1191 | */ | |
0122ec5b PZ |
1192 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1193 | lockdep_is_held(&task_rq(p)->lock))); | |
1194 | #endif | |
e2912009 PZ |
1195 | #endif |
1196 | ||
de1d7286 | 1197 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1198 | |
0c69774e | 1199 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1200 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1201 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1202 | p->se.nr_migrations++; |
ff303e66 | 1203 | perf_event_task_migrate(p); |
0c69774e | 1204 | } |
dd41f596 IM |
1205 | |
1206 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1207 | } |
1208 | ||
ac66f547 PZ |
1209 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1210 | { | |
da0c1e65 | 1211 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1212 | struct rq *src_rq, *dst_rq; |
1213 | ||
1214 | src_rq = task_rq(p); | |
1215 | dst_rq = cpu_rq(cpu); | |
1216 | ||
3ea94de1 | 1217 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1218 | deactivate_task(src_rq, p, 0); |
1219 | set_task_cpu(p, cpu); | |
1220 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1221 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1222 | check_preempt_curr(dst_rq, p, 0); |
1223 | } else { | |
1224 | /* | |
1225 | * Task isn't running anymore; make it appear like we migrated | |
1226 | * it before it went to sleep. This means on wakeup we make the | |
1227 | * previous cpu our targer instead of where it really is. | |
1228 | */ | |
1229 | p->wake_cpu = cpu; | |
1230 | } | |
1231 | } | |
1232 | ||
1233 | struct migration_swap_arg { | |
1234 | struct task_struct *src_task, *dst_task; | |
1235 | int src_cpu, dst_cpu; | |
1236 | }; | |
1237 | ||
1238 | static int migrate_swap_stop(void *data) | |
1239 | { | |
1240 | struct migration_swap_arg *arg = data; | |
1241 | struct rq *src_rq, *dst_rq; | |
1242 | int ret = -EAGAIN; | |
1243 | ||
62694cd5 PZ |
1244 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1245 | return -EAGAIN; | |
1246 | ||
ac66f547 PZ |
1247 | src_rq = cpu_rq(arg->src_cpu); |
1248 | dst_rq = cpu_rq(arg->dst_cpu); | |
1249 | ||
74602315 PZ |
1250 | double_raw_lock(&arg->src_task->pi_lock, |
1251 | &arg->dst_task->pi_lock); | |
ac66f547 | 1252 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1253 | |
ac66f547 PZ |
1254 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1255 | goto unlock; | |
1256 | ||
1257 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1258 | goto unlock; | |
1259 | ||
1260 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1261 | goto unlock; | |
1262 | ||
1263 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1264 | goto unlock; | |
1265 | ||
1266 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1267 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1268 | ||
1269 | ret = 0; | |
1270 | ||
1271 | unlock: | |
1272 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1273 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1274 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1275 | |
1276 | return ret; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Cross migrate two tasks | |
1281 | */ | |
1282 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1283 | { | |
1284 | struct migration_swap_arg arg; | |
1285 | int ret = -EINVAL; | |
1286 | ||
ac66f547 PZ |
1287 | arg = (struct migration_swap_arg){ |
1288 | .src_task = cur, | |
1289 | .src_cpu = task_cpu(cur), | |
1290 | .dst_task = p, | |
1291 | .dst_cpu = task_cpu(p), | |
1292 | }; | |
1293 | ||
1294 | if (arg.src_cpu == arg.dst_cpu) | |
1295 | goto out; | |
1296 | ||
6acce3ef PZ |
1297 | /* |
1298 | * These three tests are all lockless; this is OK since all of them | |
1299 | * will be re-checked with proper locks held further down the line. | |
1300 | */ | |
ac66f547 PZ |
1301 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1302 | goto out; | |
1303 | ||
1304 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1305 | goto out; | |
1306 | ||
1307 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1308 | goto out; | |
1309 | ||
286549dc | 1310 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1311 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1312 | ||
1313 | out: | |
ac66f547 PZ |
1314 | return ret; |
1315 | } | |
1316 | ||
1da177e4 LT |
1317 | /* |
1318 | * wait_task_inactive - wait for a thread to unschedule. | |
1319 | * | |
85ba2d86 RM |
1320 | * If @match_state is nonzero, it's the @p->state value just checked and |
1321 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1322 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1323 | * we return a positive number (its total switch count). If a second call | |
1324 | * a short while later returns the same number, the caller can be sure that | |
1325 | * @p has remained unscheduled the whole time. | |
1326 | * | |
1da177e4 LT |
1327 | * The caller must ensure that the task *will* unschedule sometime soon, |
1328 | * else this function might spin for a *long* time. This function can't | |
1329 | * be called with interrupts off, or it may introduce deadlock with | |
1330 | * smp_call_function() if an IPI is sent by the same process we are | |
1331 | * waiting to become inactive. | |
1332 | */ | |
85ba2d86 | 1333 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1334 | { |
1335 | unsigned long flags; | |
da0c1e65 | 1336 | int running, queued; |
85ba2d86 | 1337 | unsigned long ncsw; |
70b97a7f | 1338 | struct rq *rq; |
1da177e4 | 1339 | |
3a5c359a AK |
1340 | for (;;) { |
1341 | /* | |
1342 | * We do the initial early heuristics without holding | |
1343 | * any task-queue locks at all. We'll only try to get | |
1344 | * the runqueue lock when things look like they will | |
1345 | * work out! | |
1346 | */ | |
1347 | rq = task_rq(p); | |
fa490cfd | 1348 | |
3a5c359a AK |
1349 | /* |
1350 | * If the task is actively running on another CPU | |
1351 | * still, just relax and busy-wait without holding | |
1352 | * any locks. | |
1353 | * | |
1354 | * NOTE! Since we don't hold any locks, it's not | |
1355 | * even sure that "rq" stays as the right runqueue! | |
1356 | * But we don't care, since "task_running()" will | |
1357 | * return false if the runqueue has changed and p | |
1358 | * is actually now running somewhere else! | |
1359 | */ | |
85ba2d86 RM |
1360 | while (task_running(rq, p)) { |
1361 | if (match_state && unlikely(p->state != match_state)) | |
1362 | return 0; | |
3a5c359a | 1363 | cpu_relax(); |
85ba2d86 | 1364 | } |
fa490cfd | 1365 | |
3a5c359a AK |
1366 | /* |
1367 | * Ok, time to look more closely! We need the rq | |
1368 | * lock now, to be *sure*. If we're wrong, we'll | |
1369 | * just go back and repeat. | |
1370 | */ | |
1371 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1372 | trace_sched_wait_task(p); |
3a5c359a | 1373 | running = task_running(rq, p); |
da0c1e65 | 1374 | queued = task_on_rq_queued(p); |
85ba2d86 | 1375 | ncsw = 0; |
f31e11d8 | 1376 | if (!match_state || p->state == match_state) |
93dcf55f | 1377 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1378 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1379 | |
85ba2d86 RM |
1380 | /* |
1381 | * If it changed from the expected state, bail out now. | |
1382 | */ | |
1383 | if (unlikely(!ncsw)) | |
1384 | break; | |
1385 | ||
3a5c359a AK |
1386 | /* |
1387 | * Was it really running after all now that we | |
1388 | * checked with the proper locks actually held? | |
1389 | * | |
1390 | * Oops. Go back and try again.. | |
1391 | */ | |
1392 | if (unlikely(running)) { | |
1393 | cpu_relax(); | |
1394 | continue; | |
1395 | } | |
fa490cfd | 1396 | |
3a5c359a AK |
1397 | /* |
1398 | * It's not enough that it's not actively running, | |
1399 | * it must be off the runqueue _entirely_, and not | |
1400 | * preempted! | |
1401 | * | |
80dd99b3 | 1402 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1403 | * running right now), it's preempted, and we should |
1404 | * yield - it could be a while. | |
1405 | */ | |
da0c1e65 | 1406 | if (unlikely(queued)) { |
8eb90c30 TG |
1407 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1408 | ||
1409 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1410 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1411 | continue; |
1412 | } | |
fa490cfd | 1413 | |
3a5c359a AK |
1414 | /* |
1415 | * Ahh, all good. It wasn't running, and it wasn't | |
1416 | * runnable, which means that it will never become | |
1417 | * running in the future either. We're all done! | |
1418 | */ | |
1419 | break; | |
1420 | } | |
85ba2d86 RM |
1421 | |
1422 | return ncsw; | |
1da177e4 LT |
1423 | } |
1424 | ||
1425 | /*** | |
1426 | * kick_process - kick a running thread to enter/exit the kernel | |
1427 | * @p: the to-be-kicked thread | |
1428 | * | |
1429 | * Cause a process which is running on another CPU to enter | |
1430 | * kernel-mode, without any delay. (to get signals handled.) | |
1431 | * | |
25985edc | 1432 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1433 | * because all it wants to ensure is that the remote task enters |
1434 | * the kernel. If the IPI races and the task has been migrated | |
1435 | * to another CPU then no harm is done and the purpose has been | |
1436 | * achieved as well. | |
1437 | */ | |
36c8b586 | 1438 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1439 | { |
1440 | int cpu; | |
1441 | ||
1442 | preempt_disable(); | |
1443 | cpu = task_cpu(p); | |
1444 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1445 | smp_send_reschedule(cpu); | |
1446 | preempt_enable(); | |
1447 | } | |
b43e3521 | 1448 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1449 | |
30da688e | 1450 | /* |
013fdb80 | 1451 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1452 | * |
1453 | * A few notes on cpu_active vs cpu_online: | |
1454 | * | |
1455 | * - cpu_active must be a subset of cpu_online | |
1456 | * | |
1457 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1458 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
1459 | * cpu isn't yet part of the sched domains, and balancing will not | |
1460 | * see it. | |
1461 | * | |
1462 | * - on cpu-down we clear cpu_active() to mask the sched domains and | |
1463 | * avoid the load balancer to place new tasks on the to be removed | |
1464 | * cpu. Existing tasks will remain running there and will be taken | |
1465 | * off. | |
1466 | * | |
1467 | * This means that fallback selection must not select !active CPUs. | |
1468 | * And can assume that any active CPU must be online. Conversely | |
1469 | * select_task_rq() below may allow selection of !active CPUs in order | |
1470 | * to satisfy the above rules. | |
30da688e | 1471 | */ |
5da9a0fb PZ |
1472 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1473 | { | |
aa00d89c TC |
1474 | int nid = cpu_to_node(cpu); |
1475 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1476 | enum { cpuset, possible, fail } state = cpuset; |
1477 | int dest_cpu; | |
5da9a0fb | 1478 | |
aa00d89c TC |
1479 | /* |
1480 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1481 | * will return -1. There is no cpu on the node, and we should | |
1482 | * select the cpu on the other node. | |
1483 | */ | |
1484 | if (nid != -1) { | |
1485 | nodemask = cpumask_of_node(nid); | |
1486 | ||
1487 | /* Look for allowed, online CPU in same node. */ | |
1488 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1489 | if (!cpu_active(dest_cpu)) |
1490 | continue; | |
1491 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1492 | return dest_cpu; | |
1493 | } | |
2baab4e9 | 1494 | } |
5da9a0fb | 1495 | |
2baab4e9 PZ |
1496 | for (;;) { |
1497 | /* Any allowed, online CPU? */ | |
e3831edd | 1498 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1499 | if (!cpu_active(dest_cpu)) |
1500 | continue; | |
1501 | goto out; | |
1502 | } | |
5da9a0fb | 1503 | |
e73e85f0 | 1504 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1505 | switch (state) { |
1506 | case cpuset: | |
e73e85f0 ON |
1507 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1508 | cpuset_cpus_allowed_fallback(p); | |
1509 | state = possible; | |
1510 | break; | |
1511 | } | |
1512 | /* fall-through */ | |
2baab4e9 PZ |
1513 | case possible: |
1514 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1515 | state = fail; | |
1516 | break; | |
1517 | ||
1518 | case fail: | |
1519 | BUG(); | |
1520 | break; | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | out: | |
1525 | if (state != cpuset) { | |
1526 | /* | |
1527 | * Don't tell them about moving exiting tasks or | |
1528 | * kernel threads (both mm NULL), since they never | |
1529 | * leave kernel. | |
1530 | */ | |
1531 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1532 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1533 | task_pid_nr(p), p->comm, cpu); |
1534 | } | |
5da9a0fb PZ |
1535 | } |
1536 | ||
1537 | return dest_cpu; | |
1538 | } | |
1539 | ||
e2912009 | 1540 | /* |
013fdb80 | 1541 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1542 | */ |
970b13ba | 1543 | static inline |
ac66f547 | 1544 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1545 | { |
cbce1a68 PZ |
1546 | lockdep_assert_held(&p->pi_lock); |
1547 | ||
6c1d9410 WL |
1548 | if (p->nr_cpus_allowed > 1) |
1549 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e9d867a6 PZI |
1550 | else |
1551 | cpu = cpumask_any(tsk_cpus_allowed(p)); | |
e2912009 PZ |
1552 | |
1553 | /* | |
1554 | * In order not to call set_task_cpu() on a blocking task we need | |
1555 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1556 | * cpu. | |
1557 | * | |
1558 | * Since this is common to all placement strategies, this lives here. | |
1559 | * | |
1560 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1561 | * not worry about this generic constraint ] | |
1562 | */ | |
fa17b507 | 1563 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1564 | !cpu_online(cpu))) |
5da9a0fb | 1565 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1566 | |
1567 | return cpu; | |
970b13ba | 1568 | } |
09a40af5 MG |
1569 | |
1570 | static void update_avg(u64 *avg, u64 sample) | |
1571 | { | |
1572 | s64 diff = sample - *avg; | |
1573 | *avg += diff >> 3; | |
1574 | } | |
25834c73 PZ |
1575 | |
1576 | #else | |
1577 | ||
1578 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1579 | const struct cpumask *new_mask, bool check) | |
1580 | { | |
1581 | return set_cpus_allowed_ptr(p, new_mask); | |
1582 | } | |
1583 | ||
5cc389bc | 1584 | #endif /* CONFIG_SMP */ |
970b13ba | 1585 | |
d7c01d27 | 1586 | static void |
b84cb5df | 1587 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1588 | { |
d7c01d27 | 1589 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1590 | struct rq *rq = this_rq(); |
1591 | ||
d7c01d27 PZ |
1592 | #ifdef CONFIG_SMP |
1593 | int this_cpu = smp_processor_id(); | |
1594 | ||
1595 | if (cpu == this_cpu) { | |
1596 | schedstat_inc(rq, ttwu_local); | |
1597 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1598 | } else { | |
1599 | struct sched_domain *sd; | |
1600 | ||
1601 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1602 | rcu_read_lock(); |
d7c01d27 PZ |
1603 | for_each_domain(this_cpu, sd) { |
1604 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1605 | schedstat_inc(sd, ttwu_wake_remote); | |
1606 | break; | |
1607 | } | |
1608 | } | |
057f3fad | 1609 | rcu_read_unlock(); |
d7c01d27 | 1610 | } |
f339b9dc PZ |
1611 | |
1612 | if (wake_flags & WF_MIGRATED) | |
1613 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1614 | ||
d7c01d27 PZ |
1615 | #endif /* CONFIG_SMP */ |
1616 | ||
1617 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1618 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1619 | |
1620 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1621 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1622 | |
d7c01d27 PZ |
1623 | #endif /* CONFIG_SCHEDSTATS */ |
1624 | } | |
1625 | ||
1de64443 | 1626 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1627 | { |
9ed3811a | 1628 | activate_task(rq, p, en_flags); |
da0c1e65 | 1629 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1630 | |
1631 | /* if a worker is waking up, notify workqueue */ | |
1632 | if (p->flags & PF_WQ_WORKER) | |
1633 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1634 | } |
1635 | ||
23f41eeb PZ |
1636 | /* |
1637 | * Mark the task runnable and perform wakeup-preemption. | |
1638 | */ | |
89363381 | 1639 | static void |
23f41eeb | 1640 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1641 | { |
9ed3811a | 1642 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1643 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1644 | trace_sched_wakeup(p); |
1645 | ||
9ed3811a | 1646 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1647 | if (p->sched_class->task_woken) { |
1648 | /* | |
cbce1a68 PZ |
1649 | * Our task @p is fully woken up and running; so its safe to |
1650 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1651 | */ |
cbce1a68 | 1652 | lockdep_unpin_lock(&rq->lock); |
9ed3811a | 1653 | p->sched_class->task_woken(rq, p); |
cbce1a68 | 1654 | lockdep_pin_lock(&rq->lock); |
4c9a4bc8 | 1655 | } |
9ed3811a | 1656 | |
e69c6341 | 1657 | if (rq->idle_stamp) { |
78becc27 | 1658 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1659 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1660 | |
abfafa54 JL |
1661 | update_avg(&rq->avg_idle, delta); |
1662 | ||
1663 | if (rq->avg_idle > max) | |
9ed3811a | 1664 | rq->avg_idle = max; |
abfafa54 | 1665 | |
9ed3811a TH |
1666 | rq->idle_stamp = 0; |
1667 | } | |
1668 | #endif | |
1669 | } | |
1670 | ||
c05fbafb PZ |
1671 | static void |
1672 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1673 | { | |
cbce1a68 PZ |
1674 | lockdep_assert_held(&rq->lock); |
1675 | ||
c05fbafb PZ |
1676 | #ifdef CONFIG_SMP |
1677 | if (p->sched_contributes_to_load) | |
1678 | rq->nr_uninterruptible--; | |
1679 | #endif | |
1680 | ||
1681 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1682 | ttwu_do_wakeup(rq, p, wake_flags); | |
1683 | } | |
1684 | ||
1685 | /* | |
1686 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1687 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1688 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1689 | * the task is still ->on_rq. | |
1690 | */ | |
1691 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1692 | { | |
1693 | struct rq *rq; | |
1694 | int ret = 0; | |
1695 | ||
1696 | rq = __task_rq_lock(p); | |
da0c1e65 | 1697 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1698 | /* check_preempt_curr() may use rq clock */ |
1699 | update_rq_clock(rq); | |
c05fbafb PZ |
1700 | ttwu_do_wakeup(rq, p, wake_flags); |
1701 | ret = 1; | |
1702 | } | |
1703 | __task_rq_unlock(rq); | |
1704 | ||
1705 | return ret; | |
1706 | } | |
1707 | ||
317f3941 | 1708 | #ifdef CONFIG_SMP |
e3baac47 | 1709 | void sched_ttwu_pending(void) |
317f3941 PZ |
1710 | { |
1711 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1712 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1713 | struct task_struct *p; | |
e3baac47 | 1714 | unsigned long flags; |
317f3941 | 1715 | |
e3baac47 PZ |
1716 | if (!llist) |
1717 | return; | |
1718 | ||
1719 | raw_spin_lock_irqsave(&rq->lock, flags); | |
cbce1a68 | 1720 | lockdep_pin_lock(&rq->lock); |
317f3941 | 1721 | |
fa14ff4a PZ |
1722 | while (llist) { |
1723 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1724 | llist = llist_next(llist); | |
317f3941 PZ |
1725 | ttwu_do_activate(rq, p, 0); |
1726 | } | |
1727 | ||
cbce1a68 | 1728 | lockdep_unpin_lock(&rq->lock); |
e3baac47 | 1729 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1730 | } |
1731 | ||
1732 | void scheduler_ipi(void) | |
1733 | { | |
f27dde8d PZ |
1734 | /* |
1735 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1736 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1737 | * this IPI. | |
1738 | */ | |
8cb75e0c | 1739 | preempt_fold_need_resched(); |
f27dde8d | 1740 | |
fd2ac4f4 | 1741 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1742 | return; |
1743 | ||
1744 | /* | |
1745 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1746 | * traditionally all their work was done from the interrupt return | |
1747 | * path. Now that we actually do some work, we need to make sure | |
1748 | * we do call them. | |
1749 | * | |
1750 | * Some archs already do call them, luckily irq_enter/exit nest | |
1751 | * properly. | |
1752 | * | |
1753 | * Arguably we should visit all archs and update all handlers, | |
1754 | * however a fair share of IPIs are still resched only so this would | |
1755 | * somewhat pessimize the simple resched case. | |
1756 | */ | |
1757 | irq_enter(); | |
fa14ff4a | 1758 | sched_ttwu_pending(); |
ca38062e SS |
1759 | |
1760 | /* | |
1761 | * Check if someone kicked us for doing the nohz idle load balance. | |
1762 | */ | |
873b4c65 | 1763 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1764 | this_rq()->idle_balance = 1; |
ca38062e | 1765 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1766 | } |
c5d753a5 | 1767 | irq_exit(); |
317f3941 PZ |
1768 | } |
1769 | ||
1770 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1771 | { | |
e3baac47 PZ |
1772 | struct rq *rq = cpu_rq(cpu); |
1773 | ||
1774 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1775 | if (!set_nr_if_polling(rq->idle)) | |
1776 | smp_send_reschedule(cpu); | |
1777 | else | |
1778 | trace_sched_wake_idle_without_ipi(cpu); | |
1779 | } | |
317f3941 | 1780 | } |
d6aa8f85 | 1781 | |
f6be8af1 CL |
1782 | void wake_up_if_idle(int cpu) |
1783 | { | |
1784 | struct rq *rq = cpu_rq(cpu); | |
1785 | unsigned long flags; | |
1786 | ||
fd7de1e8 AL |
1787 | rcu_read_lock(); |
1788 | ||
1789 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1790 | goto out; | |
f6be8af1 CL |
1791 | |
1792 | if (set_nr_if_polling(rq->idle)) { | |
1793 | trace_sched_wake_idle_without_ipi(cpu); | |
1794 | } else { | |
1795 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1796 | if (is_idle_task(rq->curr)) | |
1797 | smp_send_reschedule(cpu); | |
1798 | /* Else cpu is not in idle, do nothing here */ | |
1799 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1800 | } | |
fd7de1e8 AL |
1801 | |
1802 | out: | |
1803 | rcu_read_unlock(); | |
f6be8af1 CL |
1804 | } |
1805 | ||
39be3501 | 1806 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1807 | { |
1808 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1809 | } | |
d6aa8f85 | 1810 | #endif /* CONFIG_SMP */ |
317f3941 | 1811 | |
c05fbafb PZ |
1812 | static void ttwu_queue(struct task_struct *p, int cpu) |
1813 | { | |
1814 | struct rq *rq = cpu_rq(cpu); | |
1815 | ||
17d9f311 | 1816 | #if defined(CONFIG_SMP) |
39be3501 | 1817 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1818 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1819 | ttwu_queue_remote(p, cpu); |
1820 | return; | |
1821 | } | |
1822 | #endif | |
1823 | ||
c05fbafb | 1824 | raw_spin_lock(&rq->lock); |
cbce1a68 | 1825 | lockdep_pin_lock(&rq->lock); |
c05fbafb | 1826 | ttwu_do_activate(rq, p, 0); |
cbce1a68 | 1827 | lockdep_unpin_lock(&rq->lock); |
c05fbafb | 1828 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1829 | } |
1830 | ||
8643cda5 PZ |
1831 | /* |
1832 | * Notes on Program-Order guarantees on SMP systems. | |
1833 | * | |
1834 | * MIGRATION | |
1835 | * | |
1836 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1837 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1838 | * execution on its new cpu [c1]. | |
1839 | * | |
1840 | * For migration (of runnable tasks) this is provided by the following means: | |
1841 | * | |
1842 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1843 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1844 | * rq(c1)->lock (if not at the same time, then in that order). | |
1845 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1846 | * | |
1847 | * Transitivity guarantees that B happens after A and C after B. | |
1848 | * Note: we only require RCpc transitivity. | |
1849 | * Note: the cpu doing B need not be c0 or c1 | |
1850 | * | |
1851 | * Example: | |
1852 | * | |
1853 | * CPU0 CPU1 CPU2 | |
1854 | * | |
1855 | * LOCK rq(0)->lock | |
1856 | * sched-out X | |
1857 | * sched-in Y | |
1858 | * UNLOCK rq(0)->lock | |
1859 | * | |
1860 | * LOCK rq(0)->lock // orders against CPU0 | |
1861 | * dequeue X | |
1862 | * UNLOCK rq(0)->lock | |
1863 | * | |
1864 | * LOCK rq(1)->lock | |
1865 | * enqueue X | |
1866 | * UNLOCK rq(1)->lock | |
1867 | * | |
1868 | * LOCK rq(1)->lock // orders against CPU2 | |
1869 | * sched-out Z | |
1870 | * sched-in X | |
1871 | * UNLOCK rq(1)->lock | |
1872 | * | |
1873 | * | |
1874 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1875 | * | |
1876 | * For blocking we (obviously) need to provide the same guarantee as for | |
1877 | * migration. However the means are completely different as there is no lock | |
1878 | * chain to provide order. Instead we do: | |
1879 | * | |
1880 | * 1) smp_store_release(X->on_cpu, 0) | |
1881 | * 2) smp_cond_acquire(!X->on_cpu) | |
1882 | * | |
1883 | * Example: | |
1884 | * | |
1885 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1886 | * | |
1887 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1888 | * dequeue X | |
1889 | * sched-out X | |
1890 | * smp_store_release(X->on_cpu, 0); | |
1891 | * | |
1892 | * smp_cond_acquire(!X->on_cpu); | |
1893 | * X->state = WAKING | |
1894 | * set_task_cpu(X,2) | |
1895 | * | |
1896 | * LOCK rq(2)->lock | |
1897 | * enqueue X | |
1898 | * X->state = RUNNING | |
1899 | * UNLOCK rq(2)->lock | |
1900 | * | |
1901 | * LOCK rq(2)->lock // orders against CPU1 | |
1902 | * sched-out Z | |
1903 | * sched-in X | |
1904 | * UNLOCK rq(2)->lock | |
1905 | * | |
1906 | * UNLOCK X->pi_lock | |
1907 | * UNLOCK rq(0)->lock | |
1908 | * | |
1909 | * | |
1910 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1911 | * must observe the state that lead to our wakeup. That is, not only must our | |
1912 | * task observe its own prior state, it must also observe the stores prior to | |
1913 | * its wakeup. | |
1914 | * | |
1915 | * This means that any means of doing remote wakeups must order the CPU doing | |
1916 | * the wakeup against the CPU the task is going to end up running on. This, | |
1917 | * however, is already required for the regular Program-Order guarantee above, | |
1918 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_acquire). | |
1919 | * | |
1920 | */ | |
1921 | ||
9ed3811a | 1922 | /** |
1da177e4 | 1923 | * try_to_wake_up - wake up a thread |
9ed3811a | 1924 | * @p: the thread to be awakened |
1da177e4 | 1925 | * @state: the mask of task states that can be woken |
9ed3811a | 1926 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1927 | * |
1928 | * Put it on the run-queue if it's not already there. The "current" | |
1929 | * thread is always on the run-queue (except when the actual | |
1930 | * re-schedule is in progress), and as such you're allowed to do | |
1931 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1932 | * runnable without the overhead of this. | |
1933 | * | |
e69f6186 | 1934 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1935 | * or @state didn't match @p's state. |
1da177e4 | 1936 | */ |
e4a52bcb PZ |
1937 | static int |
1938 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1939 | { |
1da177e4 | 1940 | unsigned long flags; |
c05fbafb | 1941 | int cpu, success = 0; |
2398f2c6 | 1942 | |
e0acd0a6 ON |
1943 | /* |
1944 | * If we are going to wake up a thread waiting for CONDITION we | |
1945 | * need to ensure that CONDITION=1 done by the caller can not be | |
1946 | * reordered with p->state check below. This pairs with mb() in | |
1947 | * set_current_state() the waiting thread does. | |
1948 | */ | |
1949 | smp_mb__before_spinlock(); | |
013fdb80 | 1950 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1951 | if (!(p->state & state)) |
1da177e4 LT |
1952 | goto out; |
1953 | ||
fbd705a0 PZ |
1954 | trace_sched_waking(p); |
1955 | ||
c05fbafb | 1956 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1957 | cpu = task_cpu(p); |
1da177e4 | 1958 | |
c05fbafb PZ |
1959 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1960 | goto stat; | |
1da177e4 | 1961 | |
1da177e4 | 1962 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
1963 | /* |
1964 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
1965 | * possible to, falsely, observe p->on_cpu == 0. | |
1966 | * | |
1967 | * One must be running (->on_cpu == 1) in order to remove oneself | |
1968 | * from the runqueue. | |
1969 | * | |
1970 | * [S] ->on_cpu = 1; [L] ->on_rq | |
1971 | * UNLOCK rq->lock | |
1972 | * RMB | |
1973 | * LOCK rq->lock | |
1974 | * [S] ->on_rq = 0; [L] ->on_cpu | |
1975 | * | |
1976 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
1977 | * from the consecutive calls to schedule(); the first switching to our | |
1978 | * task, the second putting it to sleep. | |
1979 | */ | |
1980 | smp_rmb(); | |
1981 | ||
e9c84311 | 1982 | /* |
c05fbafb PZ |
1983 | * If the owning (remote) cpu is still in the middle of schedule() with |
1984 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
1985 | * |
1986 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
1987 | * | |
1988 | * This ensures that tasks getting woken will be fully ordered against | |
1989 | * their previous state and preserve Program Order. | |
0970d299 | 1990 | */ |
b3e0b1b6 | 1991 | smp_cond_acquire(!p->on_cpu); |
1da177e4 | 1992 | |
a8e4f2ea | 1993 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1994 | p->state = TASK_WAKING; |
e7693a36 | 1995 | |
e4a52bcb | 1996 | if (p->sched_class->task_waking) |
74f8e4b2 | 1997 | p->sched_class->task_waking(p); |
efbbd05a | 1998 | |
ac66f547 | 1999 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2000 | if (task_cpu(p) != cpu) { |
2001 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2002 | set_task_cpu(p, cpu); |
f339b9dc | 2003 | } |
1da177e4 | 2004 | #endif /* CONFIG_SMP */ |
1da177e4 | 2005 | |
c05fbafb PZ |
2006 | ttwu_queue(p, cpu); |
2007 | stat: | |
cb251765 MG |
2008 | if (schedstat_enabled()) |
2009 | ttwu_stat(p, cpu, wake_flags); | |
1da177e4 | 2010 | out: |
013fdb80 | 2011 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2012 | |
2013 | return success; | |
2014 | } | |
2015 | ||
21aa9af0 TH |
2016 | /** |
2017 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2018 | * @p: the thread to be awakened | |
2019 | * | |
2acca55e | 2020 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2021 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2022 | * the current task. |
21aa9af0 TH |
2023 | */ |
2024 | static void try_to_wake_up_local(struct task_struct *p) | |
2025 | { | |
2026 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2027 | |
383efcd0 TH |
2028 | if (WARN_ON_ONCE(rq != this_rq()) || |
2029 | WARN_ON_ONCE(p == current)) | |
2030 | return; | |
2031 | ||
21aa9af0 TH |
2032 | lockdep_assert_held(&rq->lock); |
2033 | ||
2acca55e | 2034 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2035 | /* |
2036 | * This is OK, because current is on_cpu, which avoids it being | |
2037 | * picked for load-balance and preemption/IRQs are still | |
2038 | * disabled avoiding further scheduler activity on it and we've | |
2039 | * not yet picked a replacement task. | |
2040 | */ | |
2041 | lockdep_unpin_lock(&rq->lock); | |
2acca55e PZ |
2042 | raw_spin_unlock(&rq->lock); |
2043 | raw_spin_lock(&p->pi_lock); | |
2044 | raw_spin_lock(&rq->lock); | |
cbce1a68 | 2045 | lockdep_pin_lock(&rq->lock); |
2acca55e PZ |
2046 | } |
2047 | ||
21aa9af0 | 2048 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2049 | goto out; |
21aa9af0 | 2050 | |
fbd705a0 PZ |
2051 | trace_sched_waking(p); |
2052 | ||
da0c1e65 | 2053 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2054 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2055 | ||
23f41eeb | 2056 | ttwu_do_wakeup(rq, p, 0); |
cb251765 MG |
2057 | if (schedstat_enabled()) |
2058 | ttwu_stat(p, smp_processor_id(), 0); | |
2acca55e PZ |
2059 | out: |
2060 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2061 | } |
2062 | ||
50fa610a DH |
2063 | /** |
2064 | * wake_up_process - Wake up a specific process | |
2065 | * @p: The process to be woken up. | |
2066 | * | |
2067 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2068 | * processes. |
2069 | * | |
2070 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2071 | * |
2072 | * It may be assumed that this function implies a write memory barrier before | |
2073 | * changing the task state if and only if any tasks are woken up. | |
2074 | */ | |
7ad5b3a5 | 2075 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2076 | { |
9067ac85 | 2077 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2078 | } |
1da177e4 LT |
2079 | EXPORT_SYMBOL(wake_up_process); |
2080 | ||
7ad5b3a5 | 2081 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2082 | { |
2083 | return try_to_wake_up(p, state, 0); | |
2084 | } | |
2085 | ||
a5e7be3b JL |
2086 | /* |
2087 | * This function clears the sched_dl_entity static params. | |
2088 | */ | |
2089 | void __dl_clear_params(struct task_struct *p) | |
2090 | { | |
2091 | struct sched_dl_entity *dl_se = &p->dl; | |
2092 | ||
2093 | dl_se->dl_runtime = 0; | |
2094 | dl_se->dl_deadline = 0; | |
2095 | dl_se->dl_period = 0; | |
2096 | dl_se->flags = 0; | |
2097 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2098 | |
2099 | dl_se->dl_throttled = 0; | |
40767b0d | 2100 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2101 | } |
2102 | ||
1da177e4 LT |
2103 | /* |
2104 | * Perform scheduler related setup for a newly forked process p. | |
2105 | * p is forked by current. | |
dd41f596 IM |
2106 | * |
2107 | * __sched_fork() is basic setup used by init_idle() too: | |
2108 | */ | |
5e1576ed | 2109 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2110 | { |
fd2f4419 PZ |
2111 | p->on_rq = 0; |
2112 | ||
2113 | p->se.on_rq = 0; | |
dd41f596 IM |
2114 | p->se.exec_start = 0; |
2115 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2116 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2117 | p->se.nr_migrations = 0; |
da7a735e | 2118 | p->se.vruntime = 0; |
fd2f4419 | 2119 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2120 | |
ad936d86 BP |
2121 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2122 | p->se.cfs_rq = NULL; | |
2123 | #endif | |
2124 | ||
6cfb0d5d | 2125 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2126 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2127 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2128 | #endif |
476d139c | 2129 | |
aab03e05 | 2130 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2131 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2132 | __dl_clear_params(p); |
aab03e05 | 2133 | |
fa717060 | 2134 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2135 | p->rt.timeout = 0; |
2136 | p->rt.time_slice = sched_rr_timeslice; | |
2137 | p->rt.on_rq = 0; | |
2138 | p->rt.on_list = 0; | |
476d139c | 2139 | |
e107be36 AK |
2140 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2141 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2142 | #endif | |
cbee9f88 PZ |
2143 | |
2144 | #ifdef CONFIG_NUMA_BALANCING | |
2145 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2146 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2147 | p->mm->numa_scan_seq = 0; |
2148 | } | |
2149 | ||
5e1576ed RR |
2150 | if (clone_flags & CLONE_VM) |
2151 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2152 | else | |
2153 | p->numa_preferred_nid = -1; | |
2154 | ||
cbee9f88 PZ |
2155 | p->node_stamp = 0ULL; |
2156 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2157 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2158 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2159 | p->numa_faults = NULL; |
7e2703e6 RR |
2160 | p->last_task_numa_placement = 0; |
2161 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2162 | |
8c8a743c | 2163 | p->numa_group = NULL; |
cbee9f88 | 2164 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2165 | } |
2166 | ||
2a595721 SD |
2167 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2168 | ||
1a687c2e | 2169 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2170 | |
1a687c2e MG |
2171 | void set_numabalancing_state(bool enabled) |
2172 | { | |
2173 | if (enabled) | |
2a595721 | 2174 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2175 | else |
2a595721 | 2176 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2177 | } |
54a43d54 AK |
2178 | |
2179 | #ifdef CONFIG_PROC_SYSCTL | |
2180 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2181 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2182 | { | |
2183 | struct ctl_table t; | |
2184 | int err; | |
2a595721 | 2185 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2186 | |
2187 | if (write && !capable(CAP_SYS_ADMIN)) | |
2188 | return -EPERM; | |
2189 | ||
2190 | t = *table; | |
2191 | t.data = &state; | |
2192 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2193 | if (err < 0) | |
2194 | return err; | |
2195 | if (write) | |
2196 | set_numabalancing_state(state); | |
2197 | return err; | |
2198 | } | |
2199 | #endif | |
2200 | #endif | |
dd41f596 | 2201 | |
cb251765 MG |
2202 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
2203 | ||
2204 | #ifdef CONFIG_SCHEDSTATS | |
2205 | static void set_schedstats(bool enabled) | |
2206 | { | |
2207 | if (enabled) | |
2208 | static_branch_enable(&sched_schedstats); | |
2209 | else | |
2210 | static_branch_disable(&sched_schedstats); | |
2211 | } | |
2212 | ||
2213 | void force_schedstat_enabled(void) | |
2214 | { | |
2215 | if (!schedstat_enabled()) { | |
2216 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2217 | static_branch_enable(&sched_schedstats); | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | static int __init setup_schedstats(char *str) | |
2222 | { | |
2223 | int ret = 0; | |
2224 | if (!str) | |
2225 | goto out; | |
2226 | ||
2227 | if (!strcmp(str, "enable")) { | |
2228 | set_schedstats(true); | |
2229 | ret = 1; | |
2230 | } else if (!strcmp(str, "disable")) { | |
2231 | set_schedstats(false); | |
2232 | ret = 1; | |
2233 | } | |
2234 | out: | |
2235 | if (!ret) | |
2236 | pr_warn("Unable to parse schedstats=\n"); | |
2237 | ||
2238 | return ret; | |
2239 | } | |
2240 | __setup("schedstats=", setup_schedstats); | |
2241 | ||
2242 | #ifdef CONFIG_PROC_SYSCTL | |
2243 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2244 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2245 | { | |
2246 | struct ctl_table t; | |
2247 | int err; | |
2248 | int state = static_branch_likely(&sched_schedstats); | |
2249 | ||
2250 | if (write && !capable(CAP_SYS_ADMIN)) | |
2251 | return -EPERM; | |
2252 | ||
2253 | t = *table; | |
2254 | t.data = &state; | |
2255 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2256 | if (err < 0) | |
2257 | return err; | |
2258 | if (write) | |
2259 | set_schedstats(state); | |
2260 | return err; | |
2261 | } | |
2262 | #endif | |
2263 | #endif | |
dd41f596 IM |
2264 | |
2265 | /* | |
2266 | * fork()/clone()-time setup: | |
2267 | */ | |
aab03e05 | 2268 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2269 | { |
0122ec5b | 2270 | unsigned long flags; |
dd41f596 IM |
2271 | int cpu = get_cpu(); |
2272 | ||
5e1576ed | 2273 | __sched_fork(clone_flags, p); |
06b83b5f | 2274 | /* |
0017d735 | 2275 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2276 | * nobody will actually run it, and a signal or other external |
2277 | * event cannot wake it up and insert it on the runqueue either. | |
2278 | */ | |
0017d735 | 2279 | p->state = TASK_RUNNING; |
dd41f596 | 2280 | |
c350a04e MG |
2281 | /* |
2282 | * Make sure we do not leak PI boosting priority to the child. | |
2283 | */ | |
2284 | p->prio = current->normal_prio; | |
2285 | ||
b9dc29e7 MG |
2286 | /* |
2287 | * Revert to default priority/policy on fork if requested. | |
2288 | */ | |
2289 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2290 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2291 | p->policy = SCHED_NORMAL; |
6c697bdf | 2292 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2293 | p->rt_priority = 0; |
2294 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2295 | p->static_prio = NICE_TO_PRIO(0); | |
2296 | ||
2297 | p->prio = p->normal_prio = __normal_prio(p); | |
2298 | set_load_weight(p); | |
6c697bdf | 2299 | |
b9dc29e7 MG |
2300 | /* |
2301 | * We don't need the reset flag anymore after the fork. It has | |
2302 | * fulfilled its duty: | |
2303 | */ | |
2304 | p->sched_reset_on_fork = 0; | |
2305 | } | |
ca94c442 | 2306 | |
aab03e05 DF |
2307 | if (dl_prio(p->prio)) { |
2308 | put_cpu(); | |
2309 | return -EAGAIN; | |
2310 | } else if (rt_prio(p->prio)) { | |
2311 | p->sched_class = &rt_sched_class; | |
2312 | } else { | |
2ddbf952 | 2313 | p->sched_class = &fair_sched_class; |
aab03e05 | 2314 | } |
b29739f9 | 2315 | |
cd29fe6f PZ |
2316 | if (p->sched_class->task_fork) |
2317 | p->sched_class->task_fork(p); | |
2318 | ||
86951599 PZ |
2319 | /* |
2320 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2321 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2322 | * is ran before sched_fork(). | |
2323 | * | |
2324 | * Silence PROVE_RCU. | |
2325 | */ | |
0122ec5b | 2326 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2327 | set_task_cpu(p, cpu); |
0122ec5b | 2328 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2329 | |
f6db8347 | 2330 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2331 | if (likely(sched_info_on())) |
52f17b6c | 2332 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2333 | #endif |
3ca7a440 PZ |
2334 | #if defined(CONFIG_SMP) |
2335 | p->on_cpu = 0; | |
4866cde0 | 2336 | #endif |
01028747 | 2337 | init_task_preempt_count(p); |
806c09a7 | 2338 | #ifdef CONFIG_SMP |
917b627d | 2339 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2340 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2341 | #endif |
917b627d | 2342 | |
476d139c | 2343 | put_cpu(); |
aab03e05 | 2344 | return 0; |
1da177e4 LT |
2345 | } |
2346 | ||
332ac17e DF |
2347 | unsigned long to_ratio(u64 period, u64 runtime) |
2348 | { | |
2349 | if (runtime == RUNTIME_INF) | |
2350 | return 1ULL << 20; | |
2351 | ||
2352 | /* | |
2353 | * Doing this here saves a lot of checks in all | |
2354 | * the calling paths, and returning zero seems | |
2355 | * safe for them anyway. | |
2356 | */ | |
2357 | if (period == 0) | |
2358 | return 0; | |
2359 | ||
2360 | return div64_u64(runtime << 20, period); | |
2361 | } | |
2362 | ||
2363 | #ifdef CONFIG_SMP | |
2364 | inline struct dl_bw *dl_bw_of(int i) | |
2365 | { | |
f78f5b90 PM |
2366 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2367 | "sched RCU must be held"); | |
332ac17e DF |
2368 | return &cpu_rq(i)->rd->dl_bw; |
2369 | } | |
2370 | ||
de212f18 | 2371 | static inline int dl_bw_cpus(int i) |
332ac17e | 2372 | { |
de212f18 PZ |
2373 | struct root_domain *rd = cpu_rq(i)->rd; |
2374 | int cpus = 0; | |
2375 | ||
f78f5b90 PM |
2376 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2377 | "sched RCU must be held"); | |
de212f18 PZ |
2378 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2379 | cpus++; | |
2380 | ||
2381 | return cpus; | |
332ac17e DF |
2382 | } |
2383 | #else | |
2384 | inline struct dl_bw *dl_bw_of(int i) | |
2385 | { | |
2386 | return &cpu_rq(i)->dl.dl_bw; | |
2387 | } | |
2388 | ||
de212f18 | 2389 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2390 | { |
2391 | return 1; | |
2392 | } | |
2393 | #endif | |
2394 | ||
332ac17e DF |
2395 | /* |
2396 | * We must be sure that accepting a new task (or allowing changing the | |
2397 | * parameters of an existing one) is consistent with the bandwidth | |
2398 | * constraints. If yes, this function also accordingly updates the currently | |
2399 | * allocated bandwidth to reflect the new situation. | |
2400 | * | |
2401 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2402 | * |
2403 | * XXX we should delay bw change until the task's 0-lag point, see | |
2404 | * __setparam_dl(). | |
332ac17e DF |
2405 | */ |
2406 | static int dl_overflow(struct task_struct *p, int policy, | |
2407 | const struct sched_attr *attr) | |
2408 | { | |
2409 | ||
2410 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2411 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2412 | u64 runtime = attr->sched_runtime; |
2413 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2414 | int cpus, err = -1; |
332ac17e DF |
2415 | |
2416 | if (new_bw == p->dl.dl_bw) | |
2417 | return 0; | |
2418 | ||
2419 | /* | |
2420 | * Either if a task, enters, leave, or stays -deadline but changes | |
2421 | * its parameters, we may need to update accordingly the total | |
2422 | * allocated bandwidth of the container. | |
2423 | */ | |
2424 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2425 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2426 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2427 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2428 | __dl_add(dl_b, new_bw); | |
2429 | err = 0; | |
2430 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2431 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2432 | __dl_clear(dl_b, p->dl.dl_bw); | |
2433 | __dl_add(dl_b, new_bw); | |
2434 | err = 0; | |
2435 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2436 | __dl_clear(dl_b, p->dl.dl_bw); | |
2437 | err = 0; | |
2438 | } | |
2439 | raw_spin_unlock(&dl_b->lock); | |
2440 | ||
2441 | return err; | |
2442 | } | |
2443 | ||
2444 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2445 | ||
1da177e4 LT |
2446 | /* |
2447 | * wake_up_new_task - wake up a newly created task for the first time. | |
2448 | * | |
2449 | * This function will do some initial scheduler statistics housekeeping | |
2450 | * that must be done for every newly created context, then puts the task | |
2451 | * on the runqueue and wakes it. | |
2452 | */ | |
3e51e3ed | 2453 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2454 | { |
2455 | unsigned long flags; | |
dd41f596 | 2456 | struct rq *rq; |
fabf318e | 2457 | |
ab2515c4 | 2458 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
98d8fd81 MR |
2459 | /* Initialize new task's runnable average */ |
2460 | init_entity_runnable_average(&p->se); | |
fabf318e PZ |
2461 | #ifdef CONFIG_SMP |
2462 | /* | |
2463 | * Fork balancing, do it here and not earlier because: | |
2464 | * - cpus_allowed can change in the fork path | |
2465 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2466 | */ |
ac66f547 | 2467 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2468 | #endif |
2469 | ||
ab2515c4 | 2470 | rq = __task_rq_lock(p); |
cd29fe6f | 2471 | activate_task(rq, p, 0); |
da0c1e65 | 2472 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2473 | trace_sched_wakeup_new(p); |
a7558e01 | 2474 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2475 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2476 | if (p->sched_class->task_woken) { |
2477 | /* | |
2478 | * Nothing relies on rq->lock after this, so its fine to | |
2479 | * drop it. | |
2480 | */ | |
2481 | lockdep_unpin_lock(&rq->lock); | |
efbbd05a | 2482 | p->sched_class->task_woken(rq, p); |
0aaafaab PZ |
2483 | lockdep_pin_lock(&rq->lock); |
2484 | } | |
9a897c5a | 2485 | #endif |
0122ec5b | 2486 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2487 | } |
2488 | ||
e107be36 AK |
2489 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2490 | ||
1cde2930 PZ |
2491 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2492 | ||
2ecd9d29 PZ |
2493 | void preempt_notifier_inc(void) |
2494 | { | |
2495 | static_key_slow_inc(&preempt_notifier_key); | |
2496 | } | |
2497 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2498 | ||
2499 | void preempt_notifier_dec(void) | |
2500 | { | |
2501 | static_key_slow_dec(&preempt_notifier_key); | |
2502 | } | |
2503 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2504 | ||
e107be36 | 2505 | /** |
80dd99b3 | 2506 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2507 | * @notifier: notifier struct to register |
e107be36 AK |
2508 | */ |
2509 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2510 | { | |
2ecd9d29 PZ |
2511 | if (!static_key_false(&preempt_notifier_key)) |
2512 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2513 | ||
e107be36 AK |
2514 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2515 | } | |
2516 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2517 | ||
2518 | /** | |
2519 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2520 | * @notifier: notifier struct to unregister |
e107be36 | 2521 | * |
d84525a8 | 2522 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2523 | */ |
2524 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2525 | { | |
2526 | hlist_del(¬ifier->link); | |
2527 | } | |
2528 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2529 | ||
1cde2930 | 2530 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2531 | { |
2532 | struct preempt_notifier *notifier; | |
e107be36 | 2533 | |
b67bfe0d | 2534 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2535 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2536 | } | |
2537 | ||
1cde2930 PZ |
2538 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2539 | { | |
2540 | if (static_key_false(&preempt_notifier_key)) | |
2541 | __fire_sched_in_preempt_notifiers(curr); | |
2542 | } | |
2543 | ||
e107be36 | 2544 | static void |
1cde2930 PZ |
2545 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2546 | struct task_struct *next) | |
e107be36 AK |
2547 | { |
2548 | struct preempt_notifier *notifier; | |
e107be36 | 2549 | |
b67bfe0d | 2550 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2551 | notifier->ops->sched_out(notifier, next); |
2552 | } | |
2553 | ||
1cde2930 PZ |
2554 | static __always_inline void |
2555 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2556 | struct task_struct *next) | |
2557 | { | |
2558 | if (static_key_false(&preempt_notifier_key)) | |
2559 | __fire_sched_out_preempt_notifiers(curr, next); | |
2560 | } | |
2561 | ||
6d6bc0ad | 2562 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2563 | |
1cde2930 | 2564 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2565 | { |
2566 | } | |
2567 | ||
1cde2930 | 2568 | static inline void |
e107be36 AK |
2569 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2570 | struct task_struct *next) | |
2571 | { | |
2572 | } | |
2573 | ||
6d6bc0ad | 2574 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2575 | |
4866cde0 NP |
2576 | /** |
2577 | * prepare_task_switch - prepare to switch tasks | |
2578 | * @rq: the runqueue preparing to switch | |
421cee29 | 2579 | * @prev: the current task that is being switched out |
4866cde0 NP |
2580 | * @next: the task we are going to switch to. |
2581 | * | |
2582 | * This is called with the rq lock held and interrupts off. It must | |
2583 | * be paired with a subsequent finish_task_switch after the context | |
2584 | * switch. | |
2585 | * | |
2586 | * prepare_task_switch sets up locking and calls architecture specific | |
2587 | * hooks. | |
2588 | */ | |
e107be36 AK |
2589 | static inline void |
2590 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2591 | struct task_struct *next) | |
4866cde0 | 2592 | { |
43148951 | 2593 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2594 | perf_event_task_sched_out(prev, next); |
e107be36 | 2595 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2596 | prepare_lock_switch(rq, next); |
2597 | prepare_arch_switch(next); | |
2598 | } | |
2599 | ||
1da177e4 LT |
2600 | /** |
2601 | * finish_task_switch - clean up after a task-switch | |
2602 | * @prev: the thread we just switched away from. | |
2603 | * | |
4866cde0 NP |
2604 | * finish_task_switch must be called after the context switch, paired |
2605 | * with a prepare_task_switch call before the context switch. | |
2606 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2607 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2608 | * |
2609 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2610 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2611 | * with the lock held can cause deadlocks; see schedule() for |
2612 | * details.) | |
dfa50b60 ON |
2613 | * |
2614 | * The context switch have flipped the stack from under us and restored the | |
2615 | * local variables which were saved when this task called schedule() in the | |
2616 | * past. prev == current is still correct but we need to recalculate this_rq | |
2617 | * because prev may have moved to another CPU. | |
1da177e4 | 2618 | */ |
dfa50b60 | 2619 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2620 | __releases(rq->lock) |
2621 | { | |
dfa50b60 | 2622 | struct rq *rq = this_rq(); |
1da177e4 | 2623 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2624 | long prev_state; |
1da177e4 | 2625 | |
609ca066 PZ |
2626 | /* |
2627 | * The previous task will have left us with a preempt_count of 2 | |
2628 | * because it left us after: | |
2629 | * | |
2630 | * schedule() | |
2631 | * preempt_disable(); // 1 | |
2632 | * __schedule() | |
2633 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2634 | * | |
2635 | * Also, see FORK_PREEMPT_COUNT. | |
2636 | */ | |
e2bf1c4b PZ |
2637 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2638 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2639 | current->comm, current->pid, preempt_count())) | |
2640 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2641 | |
1da177e4 LT |
2642 | rq->prev_mm = NULL; |
2643 | ||
2644 | /* | |
2645 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2646 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2647 | * schedule one last time. The schedule call will never return, and |
2648 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2649 | * |
2650 | * We must observe prev->state before clearing prev->on_cpu (in | |
2651 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2652 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2653 | * transition, resulting in a double drop. | |
1da177e4 | 2654 | */ |
55a101f8 | 2655 | prev_state = prev->state; |
bf9fae9f | 2656 | vtime_task_switch(prev); |
a8d757ef | 2657 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2658 | finish_lock_switch(rq, prev); |
01f23e16 | 2659 | finish_arch_post_lock_switch(); |
e8fa1362 | 2660 | |
e107be36 | 2661 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2662 | if (mm) |
2663 | mmdrop(mm); | |
c394cc9f | 2664 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2665 | if (prev->sched_class->task_dead) |
2666 | prev->sched_class->task_dead(prev); | |
2667 | ||
c6fd91f0 | 2668 | /* |
2669 | * Remove function-return probe instances associated with this | |
2670 | * task and put them back on the free list. | |
9761eea8 | 2671 | */ |
c6fd91f0 | 2672 | kprobe_flush_task(prev); |
1da177e4 | 2673 | put_task_struct(prev); |
c6fd91f0 | 2674 | } |
99e5ada9 | 2675 | |
de734f89 | 2676 | tick_nohz_task_switch(); |
dfa50b60 | 2677 | return rq; |
1da177e4 LT |
2678 | } |
2679 | ||
3f029d3c GH |
2680 | #ifdef CONFIG_SMP |
2681 | ||
3f029d3c | 2682 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2683 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2684 | { |
e3fca9e7 PZ |
2685 | struct callback_head *head, *next; |
2686 | void (*func)(struct rq *rq); | |
2687 | unsigned long flags; | |
3f029d3c | 2688 | |
e3fca9e7 PZ |
2689 | raw_spin_lock_irqsave(&rq->lock, flags); |
2690 | head = rq->balance_callback; | |
2691 | rq->balance_callback = NULL; | |
2692 | while (head) { | |
2693 | func = (void (*)(struct rq *))head->func; | |
2694 | next = head->next; | |
2695 | head->next = NULL; | |
2696 | head = next; | |
3f029d3c | 2697 | |
e3fca9e7 | 2698 | func(rq); |
3f029d3c | 2699 | } |
e3fca9e7 PZ |
2700 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2701 | } | |
2702 | ||
2703 | static inline void balance_callback(struct rq *rq) | |
2704 | { | |
2705 | if (unlikely(rq->balance_callback)) | |
2706 | __balance_callback(rq); | |
3f029d3c GH |
2707 | } |
2708 | ||
2709 | #else | |
da19ab51 | 2710 | |
e3fca9e7 | 2711 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2712 | { |
1da177e4 LT |
2713 | } |
2714 | ||
3f029d3c GH |
2715 | #endif |
2716 | ||
1da177e4 LT |
2717 | /** |
2718 | * schedule_tail - first thing a freshly forked thread must call. | |
2719 | * @prev: the thread we just switched away from. | |
2720 | */ | |
722a9f92 | 2721 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2722 | __releases(rq->lock) |
2723 | { | |
1a43a14a | 2724 | struct rq *rq; |
da19ab51 | 2725 | |
609ca066 PZ |
2726 | /* |
2727 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2728 | * finish_task_switch() for details. | |
2729 | * | |
2730 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2731 | * and the preempt_enable() will end up enabling preemption (on | |
2732 | * PREEMPT_COUNT kernels). | |
2733 | */ | |
2734 | ||
dfa50b60 | 2735 | rq = finish_task_switch(prev); |
e3fca9e7 | 2736 | balance_callback(rq); |
1a43a14a | 2737 | preempt_enable(); |
70b97a7f | 2738 | |
1da177e4 | 2739 | if (current->set_child_tid) |
b488893a | 2740 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2741 | } |
2742 | ||
2743 | /* | |
dfa50b60 | 2744 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2745 | */ |
04936948 | 2746 | static __always_inline struct rq * |
70b97a7f | 2747 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2748 | struct task_struct *next) |
1da177e4 | 2749 | { |
dd41f596 | 2750 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2751 | |
e107be36 | 2752 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2753 | |
dd41f596 IM |
2754 | mm = next->mm; |
2755 | oldmm = prev->active_mm; | |
9226d125 ZA |
2756 | /* |
2757 | * For paravirt, this is coupled with an exit in switch_to to | |
2758 | * combine the page table reload and the switch backend into | |
2759 | * one hypercall. | |
2760 | */ | |
224101ed | 2761 | arch_start_context_switch(prev); |
9226d125 | 2762 | |
31915ab4 | 2763 | if (!mm) { |
1da177e4 LT |
2764 | next->active_mm = oldmm; |
2765 | atomic_inc(&oldmm->mm_count); | |
2766 | enter_lazy_tlb(oldmm, next); | |
2767 | } else | |
2768 | switch_mm(oldmm, mm, next); | |
2769 | ||
31915ab4 | 2770 | if (!prev->mm) { |
1da177e4 | 2771 | prev->active_mm = NULL; |
1da177e4 LT |
2772 | rq->prev_mm = oldmm; |
2773 | } | |
3a5f5e48 IM |
2774 | /* |
2775 | * Since the runqueue lock will be released by the next | |
2776 | * task (which is an invalid locking op but in the case | |
2777 | * of the scheduler it's an obvious special-case), so we | |
2778 | * do an early lockdep release here: | |
2779 | */ | |
cbce1a68 | 2780 | lockdep_unpin_lock(&rq->lock); |
8a25d5de | 2781 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2782 | |
2783 | /* Here we just switch the register state and the stack. */ | |
2784 | switch_to(prev, next, prev); | |
dd41f596 | 2785 | barrier(); |
dfa50b60 ON |
2786 | |
2787 | return finish_task_switch(prev); | |
1da177e4 LT |
2788 | } |
2789 | ||
2790 | /* | |
1c3e8264 | 2791 | * nr_running and nr_context_switches: |
1da177e4 LT |
2792 | * |
2793 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2794 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2795 | */ |
2796 | unsigned long nr_running(void) | |
2797 | { | |
2798 | unsigned long i, sum = 0; | |
2799 | ||
2800 | for_each_online_cpu(i) | |
2801 | sum += cpu_rq(i)->nr_running; | |
2802 | ||
2803 | return sum; | |
f711f609 | 2804 | } |
1da177e4 | 2805 | |
2ee507c4 TC |
2806 | /* |
2807 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2808 | * |
2809 | * Caution: this function does not check that the caller has disabled | |
2810 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2811 | * race. The caller is responsible to use it correctly, for example: | |
2812 | * | |
2813 | * - from a non-preemptable section (of course) | |
2814 | * | |
2815 | * - from a thread that is bound to a single CPU | |
2816 | * | |
2817 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2818 | */ |
2819 | bool single_task_running(void) | |
2820 | { | |
00cc1633 | 2821 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2822 | } |
2823 | EXPORT_SYMBOL(single_task_running); | |
2824 | ||
1da177e4 | 2825 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2826 | { |
cc94abfc SR |
2827 | int i; |
2828 | unsigned long long sum = 0; | |
46cb4b7c | 2829 | |
0a945022 | 2830 | for_each_possible_cpu(i) |
1da177e4 | 2831 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2832 | |
1da177e4 LT |
2833 | return sum; |
2834 | } | |
483b4ee6 | 2835 | |
1da177e4 LT |
2836 | unsigned long nr_iowait(void) |
2837 | { | |
2838 | unsigned long i, sum = 0; | |
483b4ee6 | 2839 | |
0a945022 | 2840 | for_each_possible_cpu(i) |
1da177e4 | 2841 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2842 | |
1da177e4 LT |
2843 | return sum; |
2844 | } | |
483b4ee6 | 2845 | |
8c215bd3 | 2846 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2847 | { |
8c215bd3 | 2848 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2849 | return atomic_read(&this->nr_iowait); |
2850 | } | |
46cb4b7c | 2851 | |
372ba8cb MG |
2852 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2853 | { | |
3289bdb4 PZ |
2854 | struct rq *rq = this_rq(); |
2855 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2856 | *load = rq->load.weight; | |
372ba8cb MG |
2857 | } |
2858 | ||
dd41f596 | 2859 | #ifdef CONFIG_SMP |
8a0be9ef | 2860 | |
46cb4b7c | 2861 | /* |
38022906 PZ |
2862 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2863 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2864 | */ |
38022906 | 2865 | void sched_exec(void) |
46cb4b7c | 2866 | { |
38022906 | 2867 | struct task_struct *p = current; |
1da177e4 | 2868 | unsigned long flags; |
0017d735 | 2869 | int dest_cpu; |
46cb4b7c | 2870 | |
8f42ced9 | 2871 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2872 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2873 | if (dest_cpu == smp_processor_id()) |
2874 | goto unlock; | |
38022906 | 2875 | |
8f42ced9 | 2876 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2877 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2878 | |
8f42ced9 PZ |
2879 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2880 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2881 | return; |
2882 | } | |
0017d735 | 2883 | unlock: |
8f42ced9 | 2884 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2885 | } |
dd41f596 | 2886 | |
1da177e4 LT |
2887 | #endif |
2888 | ||
1da177e4 | 2889 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2890 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2891 | |
2892 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2893 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2894 | |
c5f8d995 HS |
2895 | /* |
2896 | * Return accounted runtime for the task. | |
2897 | * In case the task is currently running, return the runtime plus current's | |
2898 | * pending runtime that have not been accounted yet. | |
2899 | */ | |
2900 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2901 | { | |
2902 | unsigned long flags; | |
2903 | struct rq *rq; | |
6e998916 | 2904 | u64 ns; |
c5f8d995 | 2905 | |
911b2898 PZ |
2906 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2907 | /* | |
2908 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2909 | * So we have a optimization chance when the task's delta_exec is 0. | |
2910 | * Reading ->on_cpu is racy, but this is ok. | |
2911 | * | |
2912 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2913 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2914 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2915 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2916 | * been accounted, so we're correct here as well. | |
911b2898 | 2917 | */ |
da0c1e65 | 2918 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2919 | return p->se.sum_exec_runtime; |
2920 | #endif | |
2921 | ||
c5f8d995 | 2922 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2923 | /* |
2924 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2925 | * project cycles that may never be accounted to this | |
2926 | * thread, breaking clock_gettime(). | |
2927 | */ | |
2928 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2929 | update_rq_clock(rq); | |
2930 | p->sched_class->update_curr(rq); | |
2931 | } | |
2932 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2933 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2934 | |
2935 | return ns; | |
2936 | } | |
48f24c4d | 2937 | |
7835b98b CL |
2938 | /* |
2939 | * This function gets called by the timer code, with HZ frequency. | |
2940 | * We call it with interrupts disabled. | |
7835b98b CL |
2941 | */ |
2942 | void scheduler_tick(void) | |
2943 | { | |
7835b98b CL |
2944 | int cpu = smp_processor_id(); |
2945 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2946 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2947 | |
2948 | sched_clock_tick(); | |
dd41f596 | 2949 | |
05fa785c | 2950 | raw_spin_lock(&rq->lock); |
3e51f33f | 2951 | update_rq_clock(rq); |
fa85ae24 | 2952 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2953 | update_cpu_load_active(rq); |
3289bdb4 | 2954 | calc_global_load_tick(rq); |
05fa785c | 2955 | raw_spin_unlock(&rq->lock); |
7835b98b | 2956 | |
e9d2b064 | 2957 | perf_event_task_tick(); |
e220d2dc | 2958 | |
e418e1c2 | 2959 | #ifdef CONFIG_SMP |
6eb57e0d | 2960 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2961 | trigger_load_balance(rq); |
e418e1c2 | 2962 | #endif |
265f22a9 | 2963 | rq_last_tick_reset(rq); |
1da177e4 LT |
2964 | } |
2965 | ||
265f22a9 FW |
2966 | #ifdef CONFIG_NO_HZ_FULL |
2967 | /** | |
2968 | * scheduler_tick_max_deferment | |
2969 | * | |
2970 | * Keep at least one tick per second when a single | |
2971 | * active task is running because the scheduler doesn't | |
2972 | * yet completely support full dynticks environment. | |
2973 | * | |
2974 | * This makes sure that uptime, CFS vruntime, load | |
2975 | * balancing, etc... continue to move forward, even | |
2976 | * with a very low granularity. | |
e69f6186 YB |
2977 | * |
2978 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2979 | */ |
2980 | u64 scheduler_tick_max_deferment(void) | |
2981 | { | |
2982 | struct rq *rq = this_rq(); | |
316c1608 | 2983 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
2984 | |
2985 | next = rq->last_sched_tick + HZ; | |
2986 | ||
2987 | if (time_before_eq(next, now)) | |
2988 | return 0; | |
2989 | ||
8fe8ff09 | 2990 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2991 | } |
265f22a9 | 2992 | #endif |
1da177e4 | 2993 | |
7e49fcce SR |
2994 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2995 | defined(CONFIG_PREEMPT_TRACER)) | |
2996 | ||
edafe3a5 | 2997 | void preempt_count_add(int val) |
1da177e4 | 2998 | { |
6cd8a4bb | 2999 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3000 | /* |
3001 | * Underflow? | |
3002 | */ | |
9a11b49a IM |
3003 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3004 | return; | |
6cd8a4bb | 3005 | #endif |
bdb43806 | 3006 | __preempt_count_add(val); |
6cd8a4bb | 3007 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3008 | /* |
3009 | * Spinlock count overflowing soon? | |
3010 | */ | |
33859f7f MOS |
3011 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3012 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3013 | #endif |
8f47b187 | 3014 | if (preempt_count() == val) { |
f904f582 | 3015 | unsigned long ip = get_lock_parent_ip(); |
8f47b187 TG |
3016 | #ifdef CONFIG_DEBUG_PREEMPT |
3017 | current->preempt_disable_ip = ip; | |
3018 | #endif | |
3019 | trace_preempt_off(CALLER_ADDR0, ip); | |
3020 | } | |
1da177e4 | 3021 | } |
bdb43806 | 3022 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3023 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3024 | |
edafe3a5 | 3025 | void preempt_count_sub(int val) |
1da177e4 | 3026 | { |
6cd8a4bb | 3027 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3028 | /* |
3029 | * Underflow? | |
3030 | */ | |
01e3eb82 | 3031 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3032 | return; |
1da177e4 LT |
3033 | /* |
3034 | * Is the spinlock portion underflowing? | |
3035 | */ | |
9a11b49a IM |
3036 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3037 | !(preempt_count() & PREEMPT_MASK))) | |
3038 | return; | |
6cd8a4bb | 3039 | #endif |
9a11b49a | 3040 | |
6cd8a4bb | 3041 | if (preempt_count() == val) |
f904f582 | 3042 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); |
bdb43806 | 3043 | __preempt_count_sub(val); |
1da177e4 | 3044 | } |
bdb43806 | 3045 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3046 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
3047 | |
3048 | #endif | |
3049 | ||
3050 | /* | |
dd41f596 | 3051 | * Print scheduling while atomic bug: |
1da177e4 | 3052 | */ |
dd41f596 | 3053 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3054 | { |
664dfa65 DJ |
3055 | if (oops_in_progress) |
3056 | return; | |
3057 | ||
3df0fc5b PZ |
3058 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3059 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3060 | |
dd41f596 | 3061 | debug_show_held_locks(prev); |
e21f5b15 | 3062 | print_modules(); |
dd41f596 IM |
3063 | if (irqs_disabled()) |
3064 | print_irqtrace_events(prev); | |
8f47b187 TG |
3065 | #ifdef CONFIG_DEBUG_PREEMPT |
3066 | if (in_atomic_preempt_off()) { | |
3067 | pr_err("Preemption disabled at:"); | |
3068 | print_ip_sym(current->preempt_disable_ip); | |
3069 | pr_cont("\n"); | |
3070 | } | |
3071 | #endif | |
6135fc1e | 3072 | dump_stack(); |
373d4d09 | 3073 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3074 | } |
1da177e4 | 3075 | |
dd41f596 IM |
3076 | /* |
3077 | * Various schedule()-time debugging checks and statistics: | |
3078 | */ | |
3079 | static inline void schedule_debug(struct task_struct *prev) | |
3080 | { | |
0d9e2632 | 3081 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
ce03e413 | 3082 | BUG_ON(task_stack_end_corrupted(prev)); |
0d9e2632 | 3083 | #endif |
b99def8b | 3084 | |
1dc0fffc | 3085 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3086 | __schedule_bug(prev); |
1dc0fffc PZ |
3087 | preempt_count_set(PREEMPT_DISABLED); |
3088 | } | |
b3fbab05 | 3089 | rcu_sleep_check(); |
dd41f596 | 3090 | |
1da177e4 LT |
3091 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3092 | ||
2d72376b | 3093 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3094 | } |
3095 | ||
3096 | /* | |
3097 | * Pick up the highest-prio task: | |
3098 | */ | |
3099 | static inline struct task_struct * | |
606dba2e | 3100 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3101 | { |
37e117c0 | 3102 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 3103 | struct task_struct *p; |
1da177e4 LT |
3104 | |
3105 | /* | |
dd41f596 IM |
3106 | * Optimization: we know that if all tasks are in |
3107 | * the fair class we can call that function directly: | |
1da177e4 | 3108 | */ |
37e117c0 | 3109 | if (likely(prev->sched_class == class && |
38033c37 | 3110 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 3111 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
3112 | if (unlikely(p == RETRY_TASK)) |
3113 | goto again; | |
3114 | ||
3115 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3116 | if (unlikely(!p)) | |
3117 | p = idle_sched_class.pick_next_task(rq, prev); | |
3118 | ||
3119 | return p; | |
1da177e4 LT |
3120 | } |
3121 | ||
37e117c0 | 3122 | again: |
34f971f6 | 3123 | for_each_class(class) { |
606dba2e | 3124 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
3125 | if (p) { |
3126 | if (unlikely(p == RETRY_TASK)) | |
3127 | goto again; | |
dd41f596 | 3128 | return p; |
37e117c0 | 3129 | } |
dd41f596 | 3130 | } |
34f971f6 PZ |
3131 | |
3132 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3133 | } |
1da177e4 | 3134 | |
dd41f596 | 3135 | /* |
c259e01a | 3136 | * __schedule() is the main scheduler function. |
edde96ea PE |
3137 | * |
3138 | * The main means of driving the scheduler and thus entering this function are: | |
3139 | * | |
3140 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3141 | * | |
3142 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3143 | * paths. For example, see arch/x86/entry_64.S. | |
3144 | * | |
3145 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3146 | * interrupt handler scheduler_tick(). | |
3147 | * | |
3148 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3149 | * task to the run-queue and that's it. | |
3150 | * | |
3151 | * Now, if the new task added to the run-queue preempts the current | |
3152 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3153 | * called on the nearest possible occasion: | |
3154 | * | |
3155 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3156 | * | |
3157 | * - in syscall or exception context, at the next outmost | |
3158 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3159 | * spin_unlock()!) | |
3160 | * | |
3161 | * - in IRQ context, return from interrupt-handler to | |
3162 | * preemptible context | |
3163 | * | |
3164 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3165 | * then at the next: | |
3166 | * | |
3167 | * - cond_resched() call | |
3168 | * - explicit schedule() call | |
3169 | * - return from syscall or exception to user-space | |
3170 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3171 | * |
b30f0e3f | 3172 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3173 | */ |
499d7955 | 3174 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3175 | { |
3176 | struct task_struct *prev, *next; | |
67ca7bde | 3177 | unsigned long *switch_count; |
dd41f596 | 3178 | struct rq *rq; |
31656519 | 3179 | int cpu; |
dd41f596 | 3180 | |
dd41f596 IM |
3181 | cpu = smp_processor_id(); |
3182 | rq = cpu_rq(cpu); | |
dd41f596 | 3183 | prev = rq->curr; |
dd41f596 | 3184 | |
b99def8b PZ |
3185 | /* |
3186 | * do_exit() calls schedule() with preemption disabled as an exception; | |
3187 | * however we must fix that up, otherwise the next task will see an | |
3188 | * inconsistent (higher) preempt count. | |
3189 | * | |
3190 | * It also avoids the below schedule_debug() test from complaining | |
3191 | * about this. | |
3192 | */ | |
3193 | if (unlikely(prev->state == TASK_DEAD)) | |
3194 | preempt_enable_no_resched_notrace(); | |
3195 | ||
dd41f596 | 3196 | schedule_debug(prev); |
1da177e4 | 3197 | |
31656519 | 3198 | if (sched_feat(HRTICK)) |
f333fdc9 | 3199 | hrtick_clear(rq); |
8f4d37ec | 3200 | |
46a5d164 PM |
3201 | local_irq_disable(); |
3202 | rcu_note_context_switch(); | |
3203 | ||
e0acd0a6 ON |
3204 | /* |
3205 | * Make sure that signal_pending_state()->signal_pending() below | |
3206 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3207 | * done by the caller to avoid the race with signal_wake_up(). | |
3208 | */ | |
3209 | smp_mb__before_spinlock(); | |
46a5d164 | 3210 | raw_spin_lock(&rq->lock); |
cbce1a68 | 3211 | lockdep_pin_lock(&rq->lock); |
1da177e4 | 3212 | |
9edfbfed PZ |
3213 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3214 | ||
246d86b5 | 3215 | switch_count = &prev->nivcsw; |
fc13aeba | 3216 | if (!preempt && prev->state) { |
21aa9af0 | 3217 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3218 | prev->state = TASK_RUNNING; |
21aa9af0 | 3219 | } else { |
2acca55e PZ |
3220 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3221 | prev->on_rq = 0; | |
3222 | ||
21aa9af0 | 3223 | /* |
2acca55e PZ |
3224 | * If a worker went to sleep, notify and ask workqueue |
3225 | * whether it wants to wake up a task to maintain | |
3226 | * concurrency. | |
21aa9af0 TH |
3227 | */ |
3228 | if (prev->flags & PF_WQ_WORKER) { | |
3229 | struct task_struct *to_wakeup; | |
3230 | ||
9b7f6597 | 3231 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 TH |
3232 | if (to_wakeup) |
3233 | try_to_wake_up_local(to_wakeup); | |
3234 | } | |
21aa9af0 | 3235 | } |
dd41f596 | 3236 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3237 | } |
3238 | ||
9edfbfed | 3239 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3240 | update_rq_clock(rq); |
3241 | ||
3242 | next = pick_next_task(rq, prev); | |
f26f9aff | 3243 | clear_tsk_need_resched(prev); |
f27dde8d | 3244 | clear_preempt_need_resched(); |
9edfbfed | 3245 | rq->clock_skip_update = 0; |
1da177e4 | 3246 | |
1da177e4 | 3247 | if (likely(prev != next)) { |
1da177e4 LT |
3248 | rq->nr_switches++; |
3249 | rq->curr = next; | |
3250 | ++*switch_count; | |
3251 | ||
c73464b1 | 3252 | trace_sched_switch(preempt, prev, next); |
dfa50b60 | 3253 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
cbce1a68 PZ |
3254 | } else { |
3255 | lockdep_unpin_lock(&rq->lock); | |
05fa785c | 3256 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3257 | } |
1da177e4 | 3258 | |
e3fca9e7 | 3259 | balance_callback(rq); |
1da177e4 | 3260 | } |
8e05e96a | 3261 | STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */ |
c259e01a | 3262 | |
9c40cef2 TG |
3263 | static inline void sched_submit_work(struct task_struct *tsk) |
3264 | { | |
3c7d5184 | 3265 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3266 | return; |
3267 | /* | |
3268 | * If we are going to sleep and we have plugged IO queued, | |
3269 | * make sure to submit it to avoid deadlocks. | |
3270 | */ | |
3271 | if (blk_needs_flush_plug(tsk)) | |
3272 | blk_schedule_flush_plug(tsk); | |
3273 | } | |
3274 | ||
722a9f92 | 3275 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3276 | { |
9c40cef2 TG |
3277 | struct task_struct *tsk = current; |
3278 | ||
3279 | sched_submit_work(tsk); | |
bfd9b2b5 | 3280 | do { |
b30f0e3f | 3281 | preempt_disable(); |
fc13aeba | 3282 | __schedule(false); |
b30f0e3f | 3283 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3284 | } while (need_resched()); |
c259e01a | 3285 | } |
1da177e4 LT |
3286 | EXPORT_SYMBOL(schedule); |
3287 | ||
91d1aa43 | 3288 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3289 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3290 | { |
3291 | /* | |
3292 | * If we come here after a random call to set_need_resched(), | |
3293 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3294 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3295 | * we find a better solution. | |
7cc78f8f AL |
3296 | * |
3297 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3298 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3299 | * too frequently to make sense yet. |
20ab65e3 | 3300 | */ |
7cc78f8f | 3301 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3302 | schedule(); |
7cc78f8f | 3303 | exception_exit(prev_state); |
20ab65e3 FW |
3304 | } |
3305 | #endif | |
3306 | ||
c5491ea7 TG |
3307 | /** |
3308 | * schedule_preempt_disabled - called with preemption disabled | |
3309 | * | |
3310 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3311 | */ | |
3312 | void __sched schedule_preempt_disabled(void) | |
3313 | { | |
ba74c144 | 3314 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3315 | schedule(); |
3316 | preempt_disable(); | |
3317 | } | |
3318 | ||
06b1f808 | 3319 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3320 | { |
3321 | do { | |
499d7955 | 3322 | preempt_disable_notrace(); |
fc13aeba | 3323 | __schedule(true); |
499d7955 | 3324 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3325 | |
3326 | /* | |
3327 | * Check again in case we missed a preemption opportunity | |
3328 | * between schedule and now. | |
3329 | */ | |
a18b5d01 FW |
3330 | } while (need_resched()); |
3331 | } | |
3332 | ||
1da177e4 LT |
3333 | #ifdef CONFIG_PREEMPT |
3334 | /* | |
2ed6e34f | 3335 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3336 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3337 | * occur there and call schedule directly. |
3338 | */ | |
722a9f92 | 3339 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3340 | { |
1da177e4 LT |
3341 | /* |
3342 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3343 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3344 | */ |
fbb00b56 | 3345 | if (likely(!preemptible())) |
1da177e4 LT |
3346 | return; |
3347 | ||
a18b5d01 | 3348 | preempt_schedule_common(); |
1da177e4 | 3349 | } |
376e2424 | 3350 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3351 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3352 | |
009f60e2 | 3353 | /** |
4eaca0a8 | 3354 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3355 | * |
3356 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3357 | * recursion and tracing preempt enabling caused by the tracing | |
3358 | * infrastructure itself. But as tracing can happen in areas coming | |
3359 | * from userspace or just about to enter userspace, a preempt enable | |
3360 | * can occur before user_exit() is called. This will cause the scheduler | |
3361 | * to be called when the system is still in usermode. | |
3362 | * | |
3363 | * To prevent this, the preempt_enable_notrace will use this function | |
3364 | * instead of preempt_schedule() to exit user context if needed before | |
3365 | * calling the scheduler. | |
3366 | */ | |
4eaca0a8 | 3367 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3368 | { |
3369 | enum ctx_state prev_ctx; | |
3370 | ||
3371 | if (likely(!preemptible())) | |
3372 | return; | |
3373 | ||
3374 | do { | |
3d8f74dd | 3375 | preempt_disable_notrace(); |
009f60e2 ON |
3376 | /* |
3377 | * Needs preempt disabled in case user_exit() is traced | |
3378 | * and the tracer calls preempt_enable_notrace() causing | |
3379 | * an infinite recursion. | |
3380 | */ | |
3381 | prev_ctx = exception_enter(); | |
fc13aeba | 3382 | __schedule(true); |
009f60e2 ON |
3383 | exception_exit(prev_ctx); |
3384 | ||
3d8f74dd | 3385 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3386 | } while (need_resched()); |
3387 | } | |
4eaca0a8 | 3388 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3389 | |
32e475d7 | 3390 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3391 | |
3392 | /* | |
2ed6e34f | 3393 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3394 | * off of irq context. |
3395 | * Note, that this is called and return with irqs disabled. This will | |
3396 | * protect us against recursive calling from irq. | |
3397 | */ | |
722a9f92 | 3398 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3399 | { |
b22366cd | 3400 | enum ctx_state prev_state; |
6478d880 | 3401 | |
2ed6e34f | 3402 | /* Catch callers which need to be fixed */ |
f27dde8d | 3403 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3404 | |
b22366cd FW |
3405 | prev_state = exception_enter(); |
3406 | ||
3a5c359a | 3407 | do { |
3d8f74dd | 3408 | preempt_disable(); |
3a5c359a | 3409 | local_irq_enable(); |
fc13aeba | 3410 | __schedule(true); |
3a5c359a | 3411 | local_irq_disable(); |
3d8f74dd | 3412 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3413 | } while (need_resched()); |
b22366cd FW |
3414 | |
3415 | exception_exit(prev_state); | |
1da177e4 LT |
3416 | } |
3417 | ||
63859d4f | 3418 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3419 | void *key) |
1da177e4 | 3420 | { |
63859d4f | 3421 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3422 | } |
1da177e4 LT |
3423 | EXPORT_SYMBOL(default_wake_function); |
3424 | ||
b29739f9 IM |
3425 | #ifdef CONFIG_RT_MUTEXES |
3426 | ||
3427 | /* | |
3428 | * rt_mutex_setprio - set the current priority of a task | |
3429 | * @p: task | |
3430 | * @prio: prio value (kernel-internal form) | |
3431 | * | |
3432 | * This function changes the 'effective' priority of a task. It does | |
3433 | * not touch ->normal_prio like __setscheduler(). | |
3434 | * | |
c365c292 TG |
3435 | * Used by the rt_mutex code to implement priority inheritance |
3436 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3437 | */ |
36c8b586 | 3438 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3439 | { |
ff77e468 | 3440 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
70b97a7f | 3441 | struct rq *rq; |
83ab0aa0 | 3442 | const struct sched_class *prev_class; |
b29739f9 | 3443 | |
aab03e05 | 3444 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3445 | |
0122ec5b | 3446 | rq = __task_rq_lock(p); |
b29739f9 | 3447 | |
1c4dd99b TG |
3448 | /* |
3449 | * Idle task boosting is a nono in general. There is one | |
3450 | * exception, when PREEMPT_RT and NOHZ is active: | |
3451 | * | |
3452 | * The idle task calls get_next_timer_interrupt() and holds | |
3453 | * the timer wheel base->lock on the CPU and another CPU wants | |
3454 | * to access the timer (probably to cancel it). We can safely | |
3455 | * ignore the boosting request, as the idle CPU runs this code | |
3456 | * with interrupts disabled and will complete the lock | |
3457 | * protected section without being interrupted. So there is no | |
3458 | * real need to boost. | |
3459 | */ | |
3460 | if (unlikely(p == rq->idle)) { | |
3461 | WARN_ON(p != rq->curr); | |
3462 | WARN_ON(p->pi_blocked_on); | |
3463 | goto out_unlock; | |
3464 | } | |
3465 | ||
a8027073 | 3466 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3467 | oldprio = p->prio; |
ff77e468 PZ |
3468 | |
3469 | if (oldprio == prio) | |
3470 | queue_flag &= ~DEQUEUE_MOVE; | |
3471 | ||
83ab0aa0 | 3472 | prev_class = p->sched_class; |
da0c1e65 | 3473 | queued = task_on_rq_queued(p); |
051a1d1a | 3474 | running = task_current(rq, p); |
da0c1e65 | 3475 | if (queued) |
ff77e468 | 3476 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3477 | if (running) |
f3cd1c4e | 3478 | put_prev_task(rq, p); |
dd41f596 | 3479 | |
2d3d891d DF |
3480 | /* |
3481 | * Boosting condition are: | |
3482 | * 1. -rt task is running and holds mutex A | |
3483 | * --> -dl task blocks on mutex A | |
3484 | * | |
3485 | * 2. -dl task is running and holds mutex A | |
3486 | * --> -dl task blocks on mutex A and could preempt the | |
3487 | * running task | |
3488 | */ | |
3489 | if (dl_prio(prio)) { | |
466af29b ON |
3490 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3491 | if (!dl_prio(p->normal_prio) || | |
3492 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3493 | p->dl.dl_boosted = 1; |
ff77e468 | 3494 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3495 | } else |
3496 | p->dl.dl_boosted = 0; | |
aab03e05 | 3497 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3498 | } else if (rt_prio(prio)) { |
3499 | if (dl_prio(oldprio)) | |
3500 | p->dl.dl_boosted = 0; | |
3501 | if (oldprio < prio) | |
ff77e468 | 3502 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3503 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3504 | } else { |
3505 | if (dl_prio(oldprio)) | |
3506 | p->dl.dl_boosted = 0; | |
746db944 BS |
3507 | if (rt_prio(oldprio)) |
3508 | p->rt.timeout = 0; | |
dd41f596 | 3509 | p->sched_class = &fair_sched_class; |
2d3d891d | 3510 | } |
dd41f596 | 3511 | |
b29739f9 IM |
3512 | p->prio = prio; |
3513 | ||
0e1f3483 HS |
3514 | if (running) |
3515 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3516 | if (queued) |
ff77e468 | 3517 | enqueue_task(rq, p, queue_flag); |
cb469845 | 3518 | |
da7a735e | 3519 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3520 | out_unlock: |
4c9a4bc8 | 3521 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3522 | __task_rq_unlock(rq); |
4c9a4bc8 PZ |
3523 | |
3524 | balance_callback(rq); | |
3525 | preempt_enable(); | |
b29739f9 | 3526 | } |
b29739f9 | 3527 | #endif |
d50dde5a | 3528 | |
36c8b586 | 3529 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3530 | { |
da0c1e65 | 3531 | int old_prio, delta, queued; |
1da177e4 | 3532 | unsigned long flags; |
70b97a7f | 3533 | struct rq *rq; |
1da177e4 | 3534 | |
75e45d51 | 3535 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3536 | return; |
3537 | /* | |
3538 | * We have to be careful, if called from sys_setpriority(), | |
3539 | * the task might be in the middle of scheduling on another CPU. | |
3540 | */ | |
3541 | rq = task_rq_lock(p, &flags); | |
3542 | /* | |
3543 | * The RT priorities are set via sched_setscheduler(), but we still | |
3544 | * allow the 'normal' nice value to be set - but as expected | |
3545 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3546 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3547 | */ |
aab03e05 | 3548 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3549 | p->static_prio = NICE_TO_PRIO(nice); |
3550 | goto out_unlock; | |
3551 | } | |
da0c1e65 KT |
3552 | queued = task_on_rq_queued(p); |
3553 | if (queued) | |
1de64443 | 3554 | dequeue_task(rq, p, DEQUEUE_SAVE); |
1da177e4 | 3555 | |
1da177e4 | 3556 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3557 | set_load_weight(p); |
b29739f9 IM |
3558 | old_prio = p->prio; |
3559 | p->prio = effective_prio(p); | |
3560 | delta = p->prio - old_prio; | |
1da177e4 | 3561 | |
da0c1e65 | 3562 | if (queued) { |
1de64443 | 3563 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3564 | /* |
d5f9f942 AM |
3565 | * If the task increased its priority or is running and |
3566 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3567 | */ |
d5f9f942 | 3568 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3569 | resched_curr(rq); |
1da177e4 LT |
3570 | } |
3571 | out_unlock: | |
0122ec5b | 3572 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3573 | } |
1da177e4 LT |
3574 | EXPORT_SYMBOL(set_user_nice); |
3575 | ||
e43379f1 MM |
3576 | /* |
3577 | * can_nice - check if a task can reduce its nice value | |
3578 | * @p: task | |
3579 | * @nice: nice value | |
3580 | */ | |
36c8b586 | 3581 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3582 | { |
024f4747 | 3583 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3584 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3585 | |
78d7d407 | 3586 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3587 | capable(CAP_SYS_NICE)); |
3588 | } | |
3589 | ||
1da177e4 LT |
3590 | #ifdef __ARCH_WANT_SYS_NICE |
3591 | ||
3592 | /* | |
3593 | * sys_nice - change the priority of the current process. | |
3594 | * @increment: priority increment | |
3595 | * | |
3596 | * sys_setpriority is a more generic, but much slower function that | |
3597 | * does similar things. | |
3598 | */ | |
5add95d4 | 3599 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3600 | { |
48f24c4d | 3601 | long nice, retval; |
1da177e4 LT |
3602 | |
3603 | /* | |
3604 | * Setpriority might change our priority at the same moment. | |
3605 | * We don't have to worry. Conceptually one call occurs first | |
3606 | * and we have a single winner. | |
3607 | */ | |
a9467fa3 | 3608 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3609 | nice = task_nice(current) + increment; |
1da177e4 | 3610 | |
a9467fa3 | 3611 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3612 | if (increment < 0 && !can_nice(current, nice)) |
3613 | return -EPERM; | |
3614 | ||
1da177e4 LT |
3615 | retval = security_task_setnice(current, nice); |
3616 | if (retval) | |
3617 | return retval; | |
3618 | ||
3619 | set_user_nice(current, nice); | |
3620 | return 0; | |
3621 | } | |
3622 | ||
3623 | #endif | |
3624 | ||
3625 | /** | |
3626 | * task_prio - return the priority value of a given task. | |
3627 | * @p: the task in question. | |
3628 | * | |
e69f6186 | 3629 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3630 | * RT tasks are offset by -200. Normal tasks are centered |
3631 | * around 0, value goes from -16 to +15. | |
3632 | */ | |
36c8b586 | 3633 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3634 | { |
3635 | return p->prio - MAX_RT_PRIO; | |
3636 | } | |
3637 | ||
1da177e4 LT |
3638 | /** |
3639 | * idle_cpu - is a given cpu idle currently? | |
3640 | * @cpu: the processor in question. | |
e69f6186 YB |
3641 | * |
3642 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3643 | */ |
3644 | int idle_cpu(int cpu) | |
3645 | { | |
908a3283 TG |
3646 | struct rq *rq = cpu_rq(cpu); |
3647 | ||
3648 | if (rq->curr != rq->idle) | |
3649 | return 0; | |
3650 | ||
3651 | if (rq->nr_running) | |
3652 | return 0; | |
3653 | ||
3654 | #ifdef CONFIG_SMP | |
3655 | if (!llist_empty(&rq->wake_list)) | |
3656 | return 0; | |
3657 | #endif | |
3658 | ||
3659 | return 1; | |
1da177e4 LT |
3660 | } |
3661 | ||
1da177e4 LT |
3662 | /** |
3663 | * idle_task - return the idle task for a given cpu. | |
3664 | * @cpu: the processor in question. | |
e69f6186 YB |
3665 | * |
3666 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3667 | */ |
36c8b586 | 3668 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3669 | { |
3670 | return cpu_rq(cpu)->idle; | |
3671 | } | |
3672 | ||
3673 | /** | |
3674 | * find_process_by_pid - find a process with a matching PID value. | |
3675 | * @pid: the pid in question. | |
e69f6186 YB |
3676 | * |
3677 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3678 | */ |
a9957449 | 3679 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3680 | { |
228ebcbe | 3681 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3682 | } |
3683 | ||
aab03e05 DF |
3684 | /* |
3685 | * This function initializes the sched_dl_entity of a newly becoming | |
3686 | * SCHED_DEADLINE task. | |
3687 | * | |
3688 | * Only the static values are considered here, the actual runtime and the | |
3689 | * absolute deadline will be properly calculated when the task is enqueued | |
3690 | * for the first time with its new policy. | |
3691 | */ | |
3692 | static void | |
3693 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3694 | { | |
3695 | struct sched_dl_entity *dl_se = &p->dl; | |
3696 | ||
aab03e05 DF |
3697 | dl_se->dl_runtime = attr->sched_runtime; |
3698 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3699 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3700 | dl_se->flags = attr->sched_flags; |
332ac17e | 3701 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3702 | |
3703 | /* | |
3704 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3705 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3706 | * | |
3707 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3708 | * point. This would include retaining the task_struct until that time | |
3709 | * and change dl_overflow() to not immediately decrement the current | |
3710 | * amount. | |
3711 | * | |
3712 | * Instead we retain the current runtime/deadline and let the new | |
3713 | * parameters take effect after the current reservation period lapses. | |
3714 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3715 | * before the current scheduling deadline. | |
3716 | * | |
3717 | * We can still have temporary overloads because we do not delay the | |
3718 | * change in bandwidth until that time; so admission control is | |
3719 | * not on the safe side. It does however guarantee tasks will never | |
3720 | * consume more than promised. | |
3721 | */ | |
aab03e05 DF |
3722 | } |
3723 | ||
c13db6b1 SR |
3724 | /* |
3725 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3726 | * it calls know not to change it. | |
3727 | */ | |
3728 | #define SETPARAM_POLICY -1 | |
3729 | ||
c365c292 TG |
3730 | static void __setscheduler_params(struct task_struct *p, |
3731 | const struct sched_attr *attr) | |
1da177e4 | 3732 | { |
d50dde5a DF |
3733 | int policy = attr->sched_policy; |
3734 | ||
c13db6b1 | 3735 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3736 | policy = p->policy; |
3737 | ||
1da177e4 | 3738 | p->policy = policy; |
d50dde5a | 3739 | |
aab03e05 DF |
3740 | if (dl_policy(policy)) |
3741 | __setparam_dl(p, attr); | |
39fd8fd2 | 3742 | else if (fair_policy(policy)) |
d50dde5a DF |
3743 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3744 | ||
39fd8fd2 PZ |
3745 | /* |
3746 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3747 | * !rt_policy. Always setting this ensures that things like | |
3748 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3749 | */ | |
3750 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3751 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3752 | set_load_weight(p); |
3753 | } | |
39fd8fd2 | 3754 | |
c365c292 TG |
3755 | /* Actually do priority change: must hold pi & rq lock. */ |
3756 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3757 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3758 | { |
3759 | __setscheduler_params(p, attr); | |
d50dde5a | 3760 | |
383afd09 | 3761 | /* |
0782e63b TG |
3762 | * Keep a potential priority boosting if called from |
3763 | * sched_setscheduler(). | |
383afd09 | 3764 | */ |
0782e63b TG |
3765 | if (keep_boost) |
3766 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3767 | else | |
3768 | p->prio = normal_prio(p); | |
383afd09 | 3769 | |
aab03e05 DF |
3770 | if (dl_prio(p->prio)) |
3771 | p->sched_class = &dl_sched_class; | |
3772 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3773 | p->sched_class = &rt_sched_class; |
3774 | else | |
3775 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3776 | } |
aab03e05 DF |
3777 | |
3778 | static void | |
3779 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3780 | { | |
3781 | struct sched_dl_entity *dl_se = &p->dl; | |
3782 | ||
3783 | attr->sched_priority = p->rt_priority; | |
3784 | attr->sched_runtime = dl_se->dl_runtime; | |
3785 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3786 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3787 | attr->sched_flags = dl_se->flags; |
3788 | } | |
3789 | ||
3790 | /* | |
3791 | * This function validates the new parameters of a -deadline task. | |
3792 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3793 | * than the runtime, as well as the period of being zero or |
332ac17e | 3794 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3795 | * user parameters are above the internal resolution of 1us (we |
3796 | * check sched_runtime only since it is always the smaller one) and | |
3797 | * below 2^63 ns (we have to check both sched_deadline and | |
3798 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3799 | */ |
3800 | static bool | |
3801 | __checkparam_dl(const struct sched_attr *attr) | |
3802 | { | |
b0827819 JL |
3803 | /* deadline != 0 */ |
3804 | if (attr->sched_deadline == 0) | |
3805 | return false; | |
3806 | ||
3807 | /* | |
3808 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3809 | * that big. | |
3810 | */ | |
3811 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3812 | return false; | |
3813 | ||
3814 | /* | |
3815 | * Since we use the MSB for wrap-around and sign issues, make | |
3816 | * sure it's not set (mind that period can be equal to zero). | |
3817 | */ | |
3818 | if (attr->sched_deadline & (1ULL << 63) || | |
3819 | attr->sched_period & (1ULL << 63)) | |
3820 | return false; | |
3821 | ||
3822 | /* runtime <= deadline <= period (if period != 0) */ | |
3823 | if ((attr->sched_period != 0 && | |
3824 | attr->sched_period < attr->sched_deadline) || | |
3825 | attr->sched_deadline < attr->sched_runtime) | |
3826 | return false; | |
3827 | ||
3828 | return true; | |
aab03e05 DF |
3829 | } |
3830 | ||
c69e8d9c DH |
3831 | /* |
3832 | * check the target process has a UID that matches the current process's | |
3833 | */ | |
3834 | static bool check_same_owner(struct task_struct *p) | |
3835 | { | |
3836 | const struct cred *cred = current_cred(), *pcred; | |
3837 | bool match; | |
3838 | ||
3839 | rcu_read_lock(); | |
3840 | pcred = __task_cred(p); | |
9c806aa0 EB |
3841 | match = (uid_eq(cred->euid, pcred->euid) || |
3842 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3843 | rcu_read_unlock(); |
3844 | return match; | |
3845 | } | |
3846 | ||
75381608 WL |
3847 | static bool dl_param_changed(struct task_struct *p, |
3848 | const struct sched_attr *attr) | |
3849 | { | |
3850 | struct sched_dl_entity *dl_se = &p->dl; | |
3851 | ||
3852 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3853 | dl_se->dl_deadline != attr->sched_deadline || | |
3854 | dl_se->dl_period != attr->sched_period || | |
3855 | dl_se->flags != attr->sched_flags) | |
3856 | return true; | |
3857 | ||
3858 | return false; | |
3859 | } | |
3860 | ||
d50dde5a DF |
3861 | static int __sched_setscheduler(struct task_struct *p, |
3862 | const struct sched_attr *attr, | |
dbc7f069 | 3863 | bool user, bool pi) |
1da177e4 | 3864 | { |
383afd09 SR |
3865 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3866 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3867 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3868 | int new_effective_prio, policy = attr->sched_policy; |
1da177e4 | 3869 | unsigned long flags; |
83ab0aa0 | 3870 | const struct sched_class *prev_class; |
70b97a7f | 3871 | struct rq *rq; |
ca94c442 | 3872 | int reset_on_fork; |
ff77e468 | 3873 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
1da177e4 | 3874 | |
66e5393a SR |
3875 | /* may grab non-irq protected spin_locks */ |
3876 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3877 | recheck: |
3878 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3879 | if (policy < 0) { |
3880 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3881 | policy = oldpolicy = p->policy; |
ca94c442 | 3882 | } else { |
7479f3c9 | 3883 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3884 | |
20f9cd2a | 3885 | if (!valid_policy(policy)) |
ca94c442 LP |
3886 | return -EINVAL; |
3887 | } | |
3888 | ||
7479f3c9 PZ |
3889 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3890 | return -EINVAL; | |
3891 | ||
1da177e4 LT |
3892 | /* |
3893 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3894 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3895 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3896 | */ |
0bb040a4 | 3897 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3898 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3899 | return -EINVAL; |
aab03e05 DF |
3900 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3901 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3902 | return -EINVAL; |
3903 | ||
37e4ab3f OC |
3904 | /* |
3905 | * Allow unprivileged RT tasks to decrease priority: | |
3906 | */ | |
961ccddd | 3907 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3908 | if (fair_policy(policy)) { |
d0ea0268 | 3909 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3910 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3911 | return -EPERM; |
3912 | } | |
3913 | ||
e05606d3 | 3914 | if (rt_policy(policy)) { |
a44702e8 ON |
3915 | unsigned long rlim_rtprio = |
3916 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3917 | |
3918 | /* can't set/change the rt policy */ | |
3919 | if (policy != p->policy && !rlim_rtprio) | |
3920 | return -EPERM; | |
3921 | ||
3922 | /* can't increase priority */ | |
d50dde5a DF |
3923 | if (attr->sched_priority > p->rt_priority && |
3924 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3925 | return -EPERM; |
3926 | } | |
c02aa73b | 3927 | |
d44753b8 JL |
3928 | /* |
3929 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3930 | * (safest behavior); in the future we would like to allow | |
3931 | * unprivileged DL tasks to increase their relative deadline | |
3932 | * or reduce their runtime (both ways reducing utilization) | |
3933 | */ | |
3934 | if (dl_policy(policy)) | |
3935 | return -EPERM; | |
3936 | ||
dd41f596 | 3937 | /* |
c02aa73b DH |
3938 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3939 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3940 | */ |
20f9cd2a | 3941 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 3942 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3943 | return -EPERM; |
3944 | } | |
5fe1d75f | 3945 | |
37e4ab3f | 3946 | /* can't change other user's priorities */ |
c69e8d9c | 3947 | if (!check_same_owner(p)) |
37e4ab3f | 3948 | return -EPERM; |
ca94c442 LP |
3949 | |
3950 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3951 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3952 | return -EPERM; | |
37e4ab3f | 3953 | } |
1da177e4 | 3954 | |
725aad24 | 3955 | if (user) { |
b0ae1981 | 3956 | retval = security_task_setscheduler(p); |
725aad24 JF |
3957 | if (retval) |
3958 | return retval; | |
3959 | } | |
3960 | ||
b29739f9 IM |
3961 | /* |
3962 | * make sure no PI-waiters arrive (or leave) while we are | |
3963 | * changing the priority of the task: | |
0122ec5b | 3964 | * |
25985edc | 3965 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3966 | * runqueue lock must be held. |
3967 | */ | |
0122ec5b | 3968 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3969 | |
34f971f6 PZ |
3970 | /* |
3971 | * Changing the policy of the stop threads its a very bad idea | |
3972 | */ | |
3973 | if (p == rq->stop) { | |
0122ec5b | 3974 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3975 | return -EINVAL; |
3976 | } | |
3977 | ||
a51e9198 | 3978 | /* |
d6b1e911 TG |
3979 | * If not changing anything there's no need to proceed further, |
3980 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3981 | */ |
d50dde5a | 3982 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3983 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3984 | goto change; |
3985 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3986 | goto change; | |
75381608 | 3987 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 3988 | goto change; |
d50dde5a | 3989 | |
d6b1e911 | 3990 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3991 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3992 | return 0; |
3993 | } | |
d50dde5a | 3994 | change: |
a51e9198 | 3995 | |
dc61b1d6 | 3996 | if (user) { |
332ac17e | 3997 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3998 | /* |
3999 | * Do not allow realtime tasks into groups that have no runtime | |
4000 | * assigned. | |
4001 | */ | |
4002 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4003 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4004 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4005 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4006 | return -EPERM; |
4007 | } | |
dc61b1d6 | 4008 | #endif |
332ac17e DF |
4009 | #ifdef CONFIG_SMP |
4010 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4011 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4012 | |
4013 | /* | |
4014 | * Don't allow tasks with an affinity mask smaller than | |
4015 | * the entire root_domain to become SCHED_DEADLINE. We | |
4016 | * will also fail if there's no bandwidth available. | |
4017 | */ | |
e4099a5e PZ |
4018 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4019 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
4020 | task_rq_unlock(rq, p, &flags); |
4021 | return -EPERM; | |
4022 | } | |
4023 | } | |
4024 | #endif | |
4025 | } | |
dc61b1d6 | 4026 | |
1da177e4 LT |
4027 | /* recheck policy now with rq lock held */ |
4028 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4029 | policy = oldpolicy = -1; | |
0122ec5b | 4030 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4031 | goto recheck; |
4032 | } | |
332ac17e DF |
4033 | |
4034 | /* | |
4035 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4036 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4037 | * is available. | |
4038 | */ | |
e4099a5e | 4039 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
4040 | task_rq_unlock(rq, p, &flags); |
4041 | return -EBUSY; | |
4042 | } | |
4043 | ||
c365c292 TG |
4044 | p->sched_reset_on_fork = reset_on_fork; |
4045 | oldprio = p->prio; | |
4046 | ||
dbc7f069 PZ |
4047 | if (pi) { |
4048 | /* | |
4049 | * Take priority boosted tasks into account. If the new | |
4050 | * effective priority is unchanged, we just store the new | |
4051 | * normal parameters and do not touch the scheduler class and | |
4052 | * the runqueue. This will be done when the task deboost | |
4053 | * itself. | |
4054 | */ | |
4055 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4056 | if (new_effective_prio == oldprio) |
4057 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4058 | } |
4059 | ||
da0c1e65 | 4060 | queued = task_on_rq_queued(p); |
051a1d1a | 4061 | running = task_current(rq, p); |
da0c1e65 | 4062 | if (queued) |
ff77e468 | 4063 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4064 | if (running) |
f3cd1c4e | 4065 | put_prev_task(rq, p); |
f6b53205 | 4066 | |
83ab0aa0 | 4067 | prev_class = p->sched_class; |
dbc7f069 | 4068 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4069 | |
0e1f3483 HS |
4070 | if (running) |
4071 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4072 | if (queued) { |
81a44c54 TG |
4073 | /* |
4074 | * We enqueue to tail when the priority of a task is | |
4075 | * increased (user space view). | |
4076 | */ | |
ff77e468 PZ |
4077 | if (oldprio < p->prio) |
4078 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4079 | |
ff77e468 | 4080 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4081 | } |
cb469845 | 4082 | |
da7a735e | 4083 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4084 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 4085 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4086 | |
dbc7f069 PZ |
4087 | if (pi) |
4088 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4089 | |
4c9a4bc8 PZ |
4090 | /* |
4091 | * Run balance callbacks after we've adjusted the PI chain. | |
4092 | */ | |
4093 | balance_callback(rq); | |
4094 | preempt_enable(); | |
95e02ca9 | 4095 | |
1da177e4 LT |
4096 | return 0; |
4097 | } | |
961ccddd | 4098 | |
7479f3c9 PZ |
4099 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4100 | const struct sched_param *param, bool check) | |
4101 | { | |
4102 | struct sched_attr attr = { | |
4103 | .sched_policy = policy, | |
4104 | .sched_priority = param->sched_priority, | |
4105 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4106 | }; | |
4107 | ||
c13db6b1 SR |
4108 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4109 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4110 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4111 | policy &= ~SCHED_RESET_ON_FORK; | |
4112 | attr.sched_policy = policy; | |
4113 | } | |
4114 | ||
dbc7f069 | 4115 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4116 | } |
961ccddd RR |
4117 | /** |
4118 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4119 | * @p: the task in question. | |
4120 | * @policy: new policy. | |
4121 | * @param: structure containing the new RT priority. | |
4122 | * | |
e69f6186 YB |
4123 | * Return: 0 on success. An error code otherwise. |
4124 | * | |
961ccddd RR |
4125 | * NOTE that the task may be already dead. |
4126 | */ | |
4127 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4128 | const struct sched_param *param) |
961ccddd | 4129 | { |
7479f3c9 | 4130 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4131 | } |
1da177e4 LT |
4132 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4133 | ||
d50dde5a DF |
4134 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4135 | { | |
dbc7f069 | 4136 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4137 | } |
4138 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4139 | ||
961ccddd RR |
4140 | /** |
4141 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4142 | * @p: the task in question. | |
4143 | * @policy: new policy. | |
4144 | * @param: structure containing the new RT priority. | |
4145 | * | |
4146 | * Just like sched_setscheduler, only don't bother checking if the | |
4147 | * current context has permission. For example, this is needed in | |
4148 | * stop_machine(): we create temporary high priority worker threads, | |
4149 | * but our caller might not have that capability. | |
e69f6186 YB |
4150 | * |
4151 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4152 | */ |
4153 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4154 | const struct sched_param *param) |
961ccddd | 4155 | { |
7479f3c9 | 4156 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4157 | } |
84778472 | 4158 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4159 | |
95cdf3b7 IM |
4160 | static int |
4161 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4162 | { |
1da177e4 LT |
4163 | struct sched_param lparam; |
4164 | struct task_struct *p; | |
36c8b586 | 4165 | int retval; |
1da177e4 LT |
4166 | |
4167 | if (!param || pid < 0) | |
4168 | return -EINVAL; | |
4169 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4170 | return -EFAULT; | |
5fe1d75f ON |
4171 | |
4172 | rcu_read_lock(); | |
4173 | retval = -ESRCH; | |
1da177e4 | 4174 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4175 | if (p != NULL) |
4176 | retval = sched_setscheduler(p, policy, &lparam); | |
4177 | rcu_read_unlock(); | |
36c8b586 | 4178 | |
1da177e4 LT |
4179 | return retval; |
4180 | } | |
4181 | ||
d50dde5a DF |
4182 | /* |
4183 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4184 | */ | |
4185 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4186 | struct sched_attr *attr) | |
4187 | { | |
4188 | u32 size; | |
4189 | int ret; | |
4190 | ||
4191 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4192 | return -EFAULT; | |
4193 | ||
4194 | /* | |
4195 | * zero the full structure, so that a short copy will be nice. | |
4196 | */ | |
4197 | memset(attr, 0, sizeof(*attr)); | |
4198 | ||
4199 | ret = get_user(size, &uattr->size); | |
4200 | if (ret) | |
4201 | return ret; | |
4202 | ||
4203 | if (size > PAGE_SIZE) /* silly large */ | |
4204 | goto err_size; | |
4205 | ||
4206 | if (!size) /* abi compat */ | |
4207 | size = SCHED_ATTR_SIZE_VER0; | |
4208 | ||
4209 | if (size < SCHED_ATTR_SIZE_VER0) | |
4210 | goto err_size; | |
4211 | ||
4212 | /* | |
4213 | * If we're handed a bigger struct than we know of, | |
4214 | * ensure all the unknown bits are 0 - i.e. new | |
4215 | * user-space does not rely on any kernel feature | |
4216 | * extensions we dont know about yet. | |
4217 | */ | |
4218 | if (size > sizeof(*attr)) { | |
4219 | unsigned char __user *addr; | |
4220 | unsigned char __user *end; | |
4221 | unsigned char val; | |
4222 | ||
4223 | addr = (void __user *)uattr + sizeof(*attr); | |
4224 | end = (void __user *)uattr + size; | |
4225 | ||
4226 | for (; addr < end; addr++) { | |
4227 | ret = get_user(val, addr); | |
4228 | if (ret) | |
4229 | return ret; | |
4230 | if (val) | |
4231 | goto err_size; | |
4232 | } | |
4233 | size = sizeof(*attr); | |
4234 | } | |
4235 | ||
4236 | ret = copy_from_user(attr, uattr, size); | |
4237 | if (ret) | |
4238 | return -EFAULT; | |
4239 | ||
4240 | /* | |
4241 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4242 | * to be strict and return an error on out-of-bounds values? | |
4243 | */ | |
75e45d51 | 4244 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4245 | |
e78c7bca | 4246 | return 0; |
d50dde5a DF |
4247 | |
4248 | err_size: | |
4249 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4250 | return -E2BIG; |
d50dde5a DF |
4251 | } |
4252 | ||
1da177e4 LT |
4253 | /** |
4254 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4255 | * @pid: the pid in question. | |
4256 | * @policy: new policy. | |
4257 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4258 | * |
4259 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4260 | */ |
5add95d4 HC |
4261 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4262 | struct sched_param __user *, param) | |
1da177e4 | 4263 | { |
c21761f1 JB |
4264 | /* negative values for policy are not valid */ |
4265 | if (policy < 0) | |
4266 | return -EINVAL; | |
4267 | ||
1da177e4 LT |
4268 | return do_sched_setscheduler(pid, policy, param); |
4269 | } | |
4270 | ||
4271 | /** | |
4272 | * sys_sched_setparam - set/change the RT priority of a thread | |
4273 | * @pid: the pid in question. | |
4274 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4275 | * |
4276 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4277 | */ |
5add95d4 | 4278 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4279 | { |
c13db6b1 | 4280 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4281 | } |
4282 | ||
d50dde5a DF |
4283 | /** |
4284 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4285 | * @pid: the pid in question. | |
5778fccf | 4286 | * @uattr: structure containing the extended parameters. |
db66d756 | 4287 | * @flags: for future extension. |
d50dde5a | 4288 | */ |
6d35ab48 PZ |
4289 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4290 | unsigned int, flags) | |
d50dde5a DF |
4291 | { |
4292 | struct sched_attr attr; | |
4293 | struct task_struct *p; | |
4294 | int retval; | |
4295 | ||
6d35ab48 | 4296 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4297 | return -EINVAL; |
4298 | ||
143cf23d MK |
4299 | retval = sched_copy_attr(uattr, &attr); |
4300 | if (retval) | |
4301 | return retval; | |
d50dde5a | 4302 | |
b14ed2c2 | 4303 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4304 | return -EINVAL; |
d50dde5a DF |
4305 | |
4306 | rcu_read_lock(); | |
4307 | retval = -ESRCH; | |
4308 | p = find_process_by_pid(pid); | |
4309 | if (p != NULL) | |
4310 | retval = sched_setattr(p, &attr); | |
4311 | rcu_read_unlock(); | |
4312 | ||
4313 | return retval; | |
4314 | } | |
4315 | ||
1da177e4 LT |
4316 | /** |
4317 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4318 | * @pid: the pid in question. | |
e69f6186 YB |
4319 | * |
4320 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4321 | * code. | |
1da177e4 | 4322 | */ |
5add95d4 | 4323 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4324 | { |
36c8b586 | 4325 | struct task_struct *p; |
3a5c359a | 4326 | int retval; |
1da177e4 LT |
4327 | |
4328 | if (pid < 0) | |
3a5c359a | 4329 | return -EINVAL; |
1da177e4 LT |
4330 | |
4331 | retval = -ESRCH; | |
5fe85be0 | 4332 | rcu_read_lock(); |
1da177e4 LT |
4333 | p = find_process_by_pid(pid); |
4334 | if (p) { | |
4335 | retval = security_task_getscheduler(p); | |
4336 | if (!retval) | |
ca94c442 LP |
4337 | retval = p->policy |
4338 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4339 | } |
5fe85be0 | 4340 | rcu_read_unlock(); |
1da177e4 LT |
4341 | return retval; |
4342 | } | |
4343 | ||
4344 | /** | |
ca94c442 | 4345 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4346 | * @pid: the pid in question. |
4347 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4348 | * |
4349 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4350 | * code. | |
1da177e4 | 4351 | */ |
5add95d4 | 4352 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4353 | { |
ce5f7f82 | 4354 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4355 | struct task_struct *p; |
3a5c359a | 4356 | int retval; |
1da177e4 LT |
4357 | |
4358 | if (!param || pid < 0) | |
3a5c359a | 4359 | return -EINVAL; |
1da177e4 | 4360 | |
5fe85be0 | 4361 | rcu_read_lock(); |
1da177e4 LT |
4362 | p = find_process_by_pid(pid); |
4363 | retval = -ESRCH; | |
4364 | if (!p) | |
4365 | goto out_unlock; | |
4366 | ||
4367 | retval = security_task_getscheduler(p); | |
4368 | if (retval) | |
4369 | goto out_unlock; | |
4370 | ||
ce5f7f82 PZ |
4371 | if (task_has_rt_policy(p)) |
4372 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4373 | rcu_read_unlock(); |
1da177e4 LT |
4374 | |
4375 | /* | |
4376 | * This one might sleep, we cannot do it with a spinlock held ... | |
4377 | */ | |
4378 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4379 | ||
1da177e4 LT |
4380 | return retval; |
4381 | ||
4382 | out_unlock: | |
5fe85be0 | 4383 | rcu_read_unlock(); |
1da177e4 LT |
4384 | return retval; |
4385 | } | |
4386 | ||
d50dde5a DF |
4387 | static int sched_read_attr(struct sched_attr __user *uattr, |
4388 | struct sched_attr *attr, | |
4389 | unsigned int usize) | |
4390 | { | |
4391 | int ret; | |
4392 | ||
4393 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4394 | return -EFAULT; | |
4395 | ||
4396 | /* | |
4397 | * If we're handed a smaller struct than we know of, | |
4398 | * ensure all the unknown bits are 0 - i.e. old | |
4399 | * user-space does not get uncomplete information. | |
4400 | */ | |
4401 | if (usize < sizeof(*attr)) { | |
4402 | unsigned char *addr; | |
4403 | unsigned char *end; | |
4404 | ||
4405 | addr = (void *)attr + usize; | |
4406 | end = (void *)attr + sizeof(*attr); | |
4407 | ||
4408 | for (; addr < end; addr++) { | |
4409 | if (*addr) | |
22400674 | 4410 | return -EFBIG; |
d50dde5a DF |
4411 | } |
4412 | ||
4413 | attr->size = usize; | |
4414 | } | |
4415 | ||
4efbc454 | 4416 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4417 | if (ret) |
4418 | return -EFAULT; | |
4419 | ||
22400674 | 4420 | return 0; |
d50dde5a DF |
4421 | } |
4422 | ||
4423 | /** | |
aab03e05 | 4424 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4425 | * @pid: the pid in question. |
5778fccf | 4426 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4427 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4428 | * @flags: for future extension. |
d50dde5a | 4429 | */ |
6d35ab48 PZ |
4430 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4431 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4432 | { |
4433 | struct sched_attr attr = { | |
4434 | .size = sizeof(struct sched_attr), | |
4435 | }; | |
4436 | struct task_struct *p; | |
4437 | int retval; | |
4438 | ||
4439 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4440 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4441 | return -EINVAL; |
4442 | ||
4443 | rcu_read_lock(); | |
4444 | p = find_process_by_pid(pid); | |
4445 | retval = -ESRCH; | |
4446 | if (!p) | |
4447 | goto out_unlock; | |
4448 | ||
4449 | retval = security_task_getscheduler(p); | |
4450 | if (retval) | |
4451 | goto out_unlock; | |
4452 | ||
4453 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4454 | if (p->sched_reset_on_fork) |
4455 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4456 | if (task_has_dl_policy(p)) |
4457 | __getparam_dl(p, &attr); | |
4458 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4459 | attr.sched_priority = p->rt_priority; |
4460 | else | |
d0ea0268 | 4461 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4462 | |
4463 | rcu_read_unlock(); | |
4464 | ||
4465 | retval = sched_read_attr(uattr, &attr, size); | |
4466 | return retval; | |
4467 | ||
4468 | out_unlock: | |
4469 | rcu_read_unlock(); | |
4470 | return retval; | |
4471 | } | |
4472 | ||
96f874e2 | 4473 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4474 | { |
5a16f3d3 | 4475 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4476 | struct task_struct *p; |
4477 | int retval; | |
1da177e4 | 4478 | |
23f5d142 | 4479 | rcu_read_lock(); |
1da177e4 LT |
4480 | |
4481 | p = find_process_by_pid(pid); | |
4482 | if (!p) { | |
23f5d142 | 4483 | rcu_read_unlock(); |
1da177e4 LT |
4484 | return -ESRCH; |
4485 | } | |
4486 | ||
23f5d142 | 4487 | /* Prevent p going away */ |
1da177e4 | 4488 | get_task_struct(p); |
23f5d142 | 4489 | rcu_read_unlock(); |
1da177e4 | 4490 | |
14a40ffc TH |
4491 | if (p->flags & PF_NO_SETAFFINITY) { |
4492 | retval = -EINVAL; | |
4493 | goto out_put_task; | |
4494 | } | |
5a16f3d3 RR |
4495 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4496 | retval = -ENOMEM; | |
4497 | goto out_put_task; | |
4498 | } | |
4499 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4500 | retval = -ENOMEM; | |
4501 | goto out_free_cpus_allowed; | |
4502 | } | |
1da177e4 | 4503 | retval = -EPERM; |
4c44aaaf EB |
4504 | if (!check_same_owner(p)) { |
4505 | rcu_read_lock(); | |
4506 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4507 | rcu_read_unlock(); | |
16303ab2 | 4508 | goto out_free_new_mask; |
4c44aaaf EB |
4509 | } |
4510 | rcu_read_unlock(); | |
4511 | } | |
1da177e4 | 4512 | |
b0ae1981 | 4513 | retval = security_task_setscheduler(p); |
e7834f8f | 4514 | if (retval) |
16303ab2 | 4515 | goto out_free_new_mask; |
e7834f8f | 4516 | |
e4099a5e PZ |
4517 | |
4518 | cpuset_cpus_allowed(p, cpus_allowed); | |
4519 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4520 | ||
332ac17e DF |
4521 | /* |
4522 | * Since bandwidth control happens on root_domain basis, | |
4523 | * if admission test is enabled, we only admit -deadline | |
4524 | * tasks allowed to run on all the CPUs in the task's | |
4525 | * root_domain. | |
4526 | */ | |
4527 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4528 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4529 | rcu_read_lock(); | |
4530 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4531 | retval = -EBUSY; |
f1e3a093 | 4532 | rcu_read_unlock(); |
16303ab2 | 4533 | goto out_free_new_mask; |
332ac17e | 4534 | } |
f1e3a093 | 4535 | rcu_read_unlock(); |
332ac17e DF |
4536 | } |
4537 | #endif | |
49246274 | 4538 | again: |
25834c73 | 4539 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4540 | |
8707d8b8 | 4541 | if (!retval) { |
5a16f3d3 RR |
4542 | cpuset_cpus_allowed(p, cpus_allowed); |
4543 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4544 | /* |
4545 | * We must have raced with a concurrent cpuset | |
4546 | * update. Just reset the cpus_allowed to the | |
4547 | * cpuset's cpus_allowed | |
4548 | */ | |
5a16f3d3 | 4549 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4550 | goto again; |
4551 | } | |
4552 | } | |
16303ab2 | 4553 | out_free_new_mask: |
5a16f3d3 RR |
4554 | free_cpumask_var(new_mask); |
4555 | out_free_cpus_allowed: | |
4556 | free_cpumask_var(cpus_allowed); | |
4557 | out_put_task: | |
1da177e4 | 4558 | put_task_struct(p); |
1da177e4 LT |
4559 | return retval; |
4560 | } | |
4561 | ||
4562 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4563 | struct cpumask *new_mask) |
1da177e4 | 4564 | { |
96f874e2 RR |
4565 | if (len < cpumask_size()) |
4566 | cpumask_clear(new_mask); | |
4567 | else if (len > cpumask_size()) | |
4568 | len = cpumask_size(); | |
4569 | ||
1da177e4 LT |
4570 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4571 | } | |
4572 | ||
4573 | /** | |
4574 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4575 | * @pid: pid of the process | |
4576 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4577 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4578 | * |
4579 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4580 | */ |
5add95d4 HC |
4581 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4582 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4583 | { |
5a16f3d3 | 4584 | cpumask_var_t new_mask; |
1da177e4 LT |
4585 | int retval; |
4586 | ||
5a16f3d3 RR |
4587 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4588 | return -ENOMEM; | |
1da177e4 | 4589 | |
5a16f3d3 RR |
4590 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4591 | if (retval == 0) | |
4592 | retval = sched_setaffinity(pid, new_mask); | |
4593 | free_cpumask_var(new_mask); | |
4594 | return retval; | |
1da177e4 LT |
4595 | } |
4596 | ||
96f874e2 | 4597 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4598 | { |
36c8b586 | 4599 | struct task_struct *p; |
31605683 | 4600 | unsigned long flags; |
1da177e4 | 4601 | int retval; |
1da177e4 | 4602 | |
23f5d142 | 4603 | rcu_read_lock(); |
1da177e4 LT |
4604 | |
4605 | retval = -ESRCH; | |
4606 | p = find_process_by_pid(pid); | |
4607 | if (!p) | |
4608 | goto out_unlock; | |
4609 | ||
e7834f8f DQ |
4610 | retval = security_task_getscheduler(p); |
4611 | if (retval) | |
4612 | goto out_unlock; | |
4613 | ||
013fdb80 | 4614 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4615 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4616 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4617 | |
4618 | out_unlock: | |
23f5d142 | 4619 | rcu_read_unlock(); |
1da177e4 | 4620 | |
9531b62f | 4621 | return retval; |
1da177e4 LT |
4622 | } |
4623 | ||
4624 | /** | |
4625 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4626 | * @pid: pid of the process | |
4627 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4628 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4629 | * |
4630 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4631 | */ |
5add95d4 HC |
4632 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4633 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4634 | { |
4635 | int ret; | |
f17c8607 | 4636 | cpumask_var_t mask; |
1da177e4 | 4637 | |
84fba5ec | 4638 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4639 | return -EINVAL; |
4640 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4641 | return -EINVAL; |
4642 | ||
f17c8607 RR |
4643 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4644 | return -ENOMEM; | |
1da177e4 | 4645 | |
f17c8607 RR |
4646 | ret = sched_getaffinity(pid, mask); |
4647 | if (ret == 0) { | |
8bc037fb | 4648 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4649 | |
4650 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4651 | ret = -EFAULT; |
4652 | else | |
cd3d8031 | 4653 | ret = retlen; |
f17c8607 RR |
4654 | } |
4655 | free_cpumask_var(mask); | |
1da177e4 | 4656 | |
f17c8607 | 4657 | return ret; |
1da177e4 LT |
4658 | } |
4659 | ||
4660 | /** | |
4661 | * sys_sched_yield - yield the current processor to other threads. | |
4662 | * | |
dd41f596 IM |
4663 | * This function yields the current CPU to other tasks. If there are no |
4664 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4665 | * |
4666 | * Return: 0. | |
1da177e4 | 4667 | */ |
5add95d4 | 4668 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4669 | { |
70b97a7f | 4670 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4671 | |
2d72376b | 4672 | schedstat_inc(rq, yld_count); |
4530d7ab | 4673 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4674 | |
4675 | /* | |
4676 | * Since we are going to call schedule() anyway, there's | |
4677 | * no need to preempt or enable interrupts: | |
4678 | */ | |
4679 | __release(rq->lock); | |
8a25d5de | 4680 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4681 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4682 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4683 | |
4684 | schedule(); | |
4685 | ||
4686 | return 0; | |
4687 | } | |
4688 | ||
02b67cc3 | 4689 | int __sched _cond_resched(void) |
1da177e4 | 4690 | { |
fe32d3cd | 4691 | if (should_resched(0)) { |
a18b5d01 | 4692 | preempt_schedule_common(); |
1da177e4 LT |
4693 | return 1; |
4694 | } | |
4695 | return 0; | |
4696 | } | |
02b67cc3 | 4697 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4698 | |
4699 | /* | |
613afbf8 | 4700 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4701 | * call schedule, and on return reacquire the lock. |
4702 | * | |
41a2d6cf | 4703 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4704 | * operations here to prevent schedule() from being called twice (once via |
4705 | * spin_unlock(), once by hand). | |
4706 | */ | |
613afbf8 | 4707 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4708 | { |
fe32d3cd | 4709 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4710 | int ret = 0; |
4711 | ||
f607c668 PZ |
4712 | lockdep_assert_held(lock); |
4713 | ||
4a81e832 | 4714 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4715 | spin_unlock(lock); |
d86ee480 | 4716 | if (resched) |
a18b5d01 | 4717 | preempt_schedule_common(); |
95c354fe NP |
4718 | else |
4719 | cpu_relax(); | |
6df3cecb | 4720 | ret = 1; |
1da177e4 | 4721 | spin_lock(lock); |
1da177e4 | 4722 | } |
6df3cecb | 4723 | return ret; |
1da177e4 | 4724 | } |
613afbf8 | 4725 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4726 | |
613afbf8 | 4727 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4728 | { |
4729 | BUG_ON(!in_softirq()); | |
4730 | ||
fe32d3cd | 4731 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4732 | local_bh_enable(); |
a18b5d01 | 4733 | preempt_schedule_common(); |
1da177e4 LT |
4734 | local_bh_disable(); |
4735 | return 1; | |
4736 | } | |
4737 | return 0; | |
4738 | } | |
613afbf8 | 4739 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4740 | |
1da177e4 LT |
4741 | /** |
4742 | * yield - yield the current processor to other threads. | |
4743 | * | |
8e3fabfd PZ |
4744 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4745 | * | |
4746 | * The scheduler is at all times free to pick the calling task as the most | |
4747 | * eligible task to run, if removing the yield() call from your code breaks | |
4748 | * it, its already broken. | |
4749 | * | |
4750 | * Typical broken usage is: | |
4751 | * | |
4752 | * while (!event) | |
4753 | * yield(); | |
4754 | * | |
4755 | * where one assumes that yield() will let 'the other' process run that will | |
4756 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4757 | * happen. Never use yield() as a progress guarantee!! | |
4758 | * | |
4759 | * If you want to use yield() to wait for something, use wait_event(). | |
4760 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4761 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4762 | */ |
4763 | void __sched yield(void) | |
4764 | { | |
4765 | set_current_state(TASK_RUNNING); | |
4766 | sys_sched_yield(); | |
4767 | } | |
1da177e4 LT |
4768 | EXPORT_SYMBOL(yield); |
4769 | ||
d95f4122 MG |
4770 | /** |
4771 | * yield_to - yield the current processor to another thread in | |
4772 | * your thread group, or accelerate that thread toward the | |
4773 | * processor it's on. | |
16addf95 RD |
4774 | * @p: target task |
4775 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4776 | * |
4777 | * It's the caller's job to ensure that the target task struct | |
4778 | * can't go away on us before we can do any checks. | |
4779 | * | |
e69f6186 | 4780 | * Return: |
7b270f60 PZ |
4781 | * true (>0) if we indeed boosted the target task. |
4782 | * false (0) if we failed to boost the target. | |
4783 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4784 | */ |
fa93384f | 4785 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4786 | { |
4787 | struct task_struct *curr = current; | |
4788 | struct rq *rq, *p_rq; | |
4789 | unsigned long flags; | |
c3c18640 | 4790 | int yielded = 0; |
d95f4122 MG |
4791 | |
4792 | local_irq_save(flags); | |
4793 | rq = this_rq(); | |
4794 | ||
4795 | again: | |
4796 | p_rq = task_rq(p); | |
7b270f60 PZ |
4797 | /* |
4798 | * If we're the only runnable task on the rq and target rq also | |
4799 | * has only one task, there's absolutely no point in yielding. | |
4800 | */ | |
4801 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4802 | yielded = -ESRCH; | |
4803 | goto out_irq; | |
4804 | } | |
4805 | ||
d95f4122 | 4806 | double_rq_lock(rq, p_rq); |
39e24d8f | 4807 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4808 | double_rq_unlock(rq, p_rq); |
4809 | goto again; | |
4810 | } | |
4811 | ||
4812 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4813 | goto out_unlock; |
d95f4122 MG |
4814 | |
4815 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4816 | goto out_unlock; |
d95f4122 MG |
4817 | |
4818 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4819 | goto out_unlock; |
d95f4122 MG |
4820 | |
4821 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4822 | if (yielded) { |
d95f4122 | 4823 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4824 | /* |
4825 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4826 | * fairness. | |
4827 | */ | |
4828 | if (preempt && rq != p_rq) | |
8875125e | 4829 | resched_curr(p_rq); |
6d1cafd8 | 4830 | } |
d95f4122 | 4831 | |
7b270f60 | 4832 | out_unlock: |
d95f4122 | 4833 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4834 | out_irq: |
d95f4122 MG |
4835 | local_irq_restore(flags); |
4836 | ||
7b270f60 | 4837 | if (yielded > 0) |
d95f4122 MG |
4838 | schedule(); |
4839 | ||
4840 | return yielded; | |
4841 | } | |
4842 | EXPORT_SYMBOL_GPL(yield_to); | |
4843 | ||
1da177e4 | 4844 | /* |
41a2d6cf | 4845 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4846 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4847 | */ |
1da177e4 LT |
4848 | long __sched io_schedule_timeout(long timeout) |
4849 | { | |
9cff8ade N |
4850 | int old_iowait = current->in_iowait; |
4851 | struct rq *rq; | |
1da177e4 LT |
4852 | long ret; |
4853 | ||
9cff8ade | 4854 | current->in_iowait = 1; |
10d784ea | 4855 | blk_schedule_flush_plug(current); |
9cff8ade | 4856 | |
0ff92245 | 4857 | delayacct_blkio_start(); |
9cff8ade | 4858 | rq = raw_rq(); |
1da177e4 LT |
4859 | atomic_inc(&rq->nr_iowait); |
4860 | ret = schedule_timeout(timeout); | |
9cff8ade | 4861 | current->in_iowait = old_iowait; |
1da177e4 | 4862 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4863 | delayacct_blkio_end(); |
9cff8ade | 4864 | |
1da177e4 LT |
4865 | return ret; |
4866 | } | |
9cff8ade | 4867 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4868 | |
4869 | /** | |
4870 | * sys_sched_get_priority_max - return maximum RT priority. | |
4871 | * @policy: scheduling class. | |
4872 | * | |
e69f6186 YB |
4873 | * Return: On success, this syscall returns the maximum |
4874 | * rt_priority that can be used by a given scheduling class. | |
4875 | * On failure, a negative error code is returned. | |
1da177e4 | 4876 | */ |
5add95d4 | 4877 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4878 | { |
4879 | int ret = -EINVAL; | |
4880 | ||
4881 | switch (policy) { | |
4882 | case SCHED_FIFO: | |
4883 | case SCHED_RR: | |
4884 | ret = MAX_USER_RT_PRIO-1; | |
4885 | break; | |
aab03e05 | 4886 | case SCHED_DEADLINE: |
1da177e4 | 4887 | case SCHED_NORMAL: |
b0a9499c | 4888 | case SCHED_BATCH: |
dd41f596 | 4889 | case SCHED_IDLE: |
1da177e4 LT |
4890 | ret = 0; |
4891 | break; | |
4892 | } | |
4893 | return ret; | |
4894 | } | |
4895 | ||
4896 | /** | |
4897 | * sys_sched_get_priority_min - return minimum RT priority. | |
4898 | * @policy: scheduling class. | |
4899 | * | |
e69f6186 YB |
4900 | * Return: On success, this syscall returns the minimum |
4901 | * rt_priority that can be used by a given scheduling class. | |
4902 | * On failure, a negative error code is returned. | |
1da177e4 | 4903 | */ |
5add95d4 | 4904 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4905 | { |
4906 | int ret = -EINVAL; | |
4907 | ||
4908 | switch (policy) { | |
4909 | case SCHED_FIFO: | |
4910 | case SCHED_RR: | |
4911 | ret = 1; | |
4912 | break; | |
aab03e05 | 4913 | case SCHED_DEADLINE: |
1da177e4 | 4914 | case SCHED_NORMAL: |
b0a9499c | 4915 | case SCHED_BATCH: |
dd41f596 | 4916 | case SCHED_IDLE: |
1da177e4 LT |
4917 | ret = 0; |
4918 | } | |
4919 | return ret; | |
4920 | } | |
4921 | ||
4922 | /** | |
4923 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4924 | * @pid: pid of the process. | |
4925 | * @interval: userspace pointer to the timeslice value. | |
4926 | * | |
4927 | * this syscall writes the default timeslice value of a given process | |
4928 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4929 | * |
4930 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4931 | * an error code. | |
1da177e4 | 4932 | */ |
17da2bd9 | 4933 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4934 | struct timespec __user *, interval) |
1da177e4 | 4935 | { |
36c8b586 | 4936 | struct task_struct *p; |
a4ec24b4 | 4937 | unsigned int time_slice; |
dba091b9 TG |
4938 | unsigned long flags; |
4939 | struct rq *rq; | |
3a5c359a | 4940 | int retval; |
1da177e4 | 4941 | struct timespec t; |
1da177e4 LT |
4942 | |
4943 | if (pid < 0) | |
3a5c359a | 4944 | return -EINVAL; |
1da177e4 LT |
4945 | |
4946 | retval = -ESRCH; | |
1a551ae7 | 4947 | rcu_read_lock(); |
1da177e4 LT |
4948 | p = find_process_by_pid(pid); |
4949 | if (!p) | |
4950 | goto out_unlock; | |
4951 | ||
4952 | retval = security_task_getscheduler(p); | |
4953 | if (retval) | |
4954 | goto out_unlock; | |
4955 | ||
dba091b9 | 4956 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4957 | time_slice = 0; |
4958 | if (p->sched_class->get_rr_interval) | |
4959 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4960 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4961 | |
1a551ae7 | 4962 | rcu_read_unlock(); |
a4ec24b4 | 4963 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4964 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4965 | return retval; |
3a5c359a | 4966 | |
1da177e4 | 4967 | out_unlock: |
1a551ae7 | 4968 | rcu_read_unlock(); |
1da177e4 LT |
4969 | return retval; |
4970 | } | |
4971 | ||
7c731e0a | 4972 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4973 | |
82a1fcb9 | 4974 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4975 | { |
1da177e4 | 4976 | unsigned long free = 0; |
4e79752c | 4977 | int ppid; |
1f8a7633 | 4978 | unsigned long state = p->state; |
1da177e4 | 4979 | |
1f8a7633 TH |
4980 | if (state) |
4981 | state = __ffs(state) + 1; | |
28d0686c | 4982 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4983 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4984 | #if BITS_PER_LONG == 32 |
1da177e4 | 4985 | if (state == TASK_RUNNING) |
3df0fc5b | 4986 | printk(KERN_CONT " running "); |
1da177e4 | 4987 | else |
3df0fc5b | 4988 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4989 | #else |
4990 | if (state == TASK_RUNNING) | |
3df0fc5b | 4991 | printk(KERN_CONT " running task "); |
1da177e4 | 4992 | else |
3df0fc5b | 4993 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4994 | #endif |
4995 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4996 | free = stack_not_used(p); |
1da177e4 | 4997 | #endif |
a90e984c | 4998 | ppid = 0; |
4e79752c | 4999 | rcu_read_lock(); |
a90e984c ON |
5000 | if (pid_alive(p)) |
5001 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5002 | rcu_read_unlock(); |
3df0fc5b | 5003 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5004 | task_pid_nr(p), ppid, |
aa47b7e0 | 5005 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5006 | |
3d1cb205 | 5007 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5008 | show_stack(p, NULL); |
1da177e4 LT |
5009 | } |
5010 | ||
e59e2ae2 | 5011 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5012 | { |
36c8b586 | 5013 | struct task_struct *g, *p; |
1da177e4 | 5014 | |
4bd77321 | 5015 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5016 | printk(KERN_INFO |
5017 | " task PC stack pid father\n"); | |
1da177e4 | 5018 | #else |
3df0fc5b PZ |
5019 | printk(KERN_INFO |
5020 | " task PC stack pid father\n"); | |
1da177e4 | 5021 | #endif |
510f5acc | 5022 | rcu_read_lock(); |
5d07f420 | 5023 | for_each_process_thread(g, p) { |
1da177e4 LT |
5024 | /* |
5025 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5026 | * console might take a lot of time: |
1da177e4 LT |
5027 | */ |
5028 | touch_nmi_watchdog(); | |
39bc89fd | 5029 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5030 | sched_show_task(p); |
5d07f420 | 5031 | } |
1da177e4 | 5032 | |
04c9167f JF |
5033 | touch_all_softlockup_watchdogs(); |
5034 | ||
dd41f596 IM |
5035 | #ifdef CONFIG_SCHED_DEBUG |
5036 | sysrq_sched_debug_show(); | |
5037 | #endif | |
510f5acc | 5038 | rcu_read_unlock(); |
e59e2ae2 IM |
5039 | /* |
5040 | * Only show locks if all tasks are dumped: | |
5041 | */ | |
93335a21 | 5042 | if (!state_filter) |
e59e2ae2 | 5043 | debug_show_all_locks(); |
1da177e4 LT |
5044 | } |
5045 | ||
0db0628d | 5046 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5047 | { |
dd41f596 | 5048 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5049 | } |
5050 | ||
f340c0d1 IM |
5051 | /** |
5052 | * init_idle - set up an idle thread for a given CPU | |
5053 | * @idle: task in question | |
5054 | * @cpu: cpu the idle task belongs to | |
5055 | * | |
5056 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5057 | * flag, to make booting more robust. | |
5058 | */ | |
0db0628d | 5059 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5060 | { |
70b97a7f | 5061 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5062 | unsigned long flags; |
5063 | ||
25834c73 PZ |
5064 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5065 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5066 | |
5e1576ed | 5067 | __sched_fork(0, idle); |
06b83b5f | 5068 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5069 | idle->se.exec_start = sched_clock(); |
5070 | ||
e1b77c92 MR |
5071 | kasan_unpoison_task_stack(idle); |
5072 | ||
de9b8f5d PZ |
5073 | #ifdef CONFIG_SMP |
5074 | /* | |
5075 | * Its possible that init_idle() gets called multiple times on a task, | |
5076 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5077 | * | |
5078 | * And since this is boot we can forgo the serialization. | |
5079 | */ | |
5080 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5081 | #endif | |
6506cf6c PZ |
5082 | /* |
5083 | * We're having a chicken and egg problem, even though we are | |
5084 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5085 | * lockdep check in task_group() will fail. | |
5086 | * | |
5087 | * Similar case to sched_fork(). / Alternatively we could | |
5088 | * use task_rq_lock() here and obtain the other rq->lock. | |
5089 | * | |
5090 | * Silence PROVE_RCU | |
5091 | */ | |
5092 | rcu_read_lock(); | |
dd41f596 | 5093 | __set_task_cpu(idle, cpu); |
6506cf6c | 5094 | rcu_read_unlock(); |
1da177e4 | 5095 | |
1da177e4 | 5096 | rq->curr = rq->idle = idle; |
da0c1e65 | 5097 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5098 | #ifdef CONFIG_SMP |
3ca7a440 | 5099 | idle->on_cpu = 1; |
4866cde0 | 5100 | #endif |
25834c73 PZ |
5101 | raw_spin_unlock(&rq->lock); |
5102 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5103 | |
5104 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5105 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5106 | |
dd41f596 IM |
5107 | /* |
5108 | * The idle tasks have their own, simple scheduling class: | |
5109 | */ | |
5110 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5111 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5112 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5113 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5114 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5115 | #endif | |
19978ca6 IM |
5116 | } |
5117 | ||
f82f8042 JL |
5118 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5119 | const struct cpumask *trial) | |
5120 | { | |
5121 | int ret = 1, trial_cpus; | |
5122 | struct dl_bw *cur_dl_b; | |
5123 | unsigned long flags; | |
5124 | ||
bb2bc55a MG |
5125 | if (!cpumask_weight(cur)) |
5126 | return ret; | |
5127 | ||
75e23e49 | 5128 | rcu_read_lock_sched(); |
f82f8042 JL |
5129 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5130 | trial_cpus = cpumask_weight(trial); | |
5131 | ||
5132 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5133 | if (cur_dl_b->bw != -1 && | |
5134 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5135 | ret = 0; | |
5136 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5137 | rcu_read_unlock_sched(); |
f82f8042 JL |
5138 | |
5139 | return ret; | |
5140 | } | |
5141 | ||
7f51412a JL |
5142 | int task_can_attach(struct task_struct *p, |
5143 | const struct cpumask *cs_cpus_allowed) | |
5144 | { | |
5145 | int ret = 0; | |
5146 | ||
5147 | /* | |
5148 | * Kthreads which disallow setaffinity shouldn't be moved | |
5149 | * to a new cpuset; we don't want to change their cpu | |
5150 | * affinity and isolating such threads by their set of | |
5151 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5152 | * applicable for such threads. This prevents checking for | |
5153 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5154 | * before cpus_allowed may be changed. | |
5155 | */ | |
5156 | if (p->flags & PF_NO_SETAFFINITY) { | |
5157 | ret = -EINVAL; | |
5158 | goto out; | |
5159 | } | |
5160 | ||
5161 | #ifdef CONFIG_SMP | |
5162 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5163 | cs_cpus_allowed)) { | |
5164 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5165 | cs_cpus_allowed); | |
75e23e49 | 5166 | struct dl_bw *dl_b; |
7f51412a JL |
5167 | bool overflow; |
5168 | int cpus; | |
5169 | unsigned long flags; | |
5170 | ||
75e23e49 JL |
5171 | rcu_read_lock_sched(); |
5172 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5173 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5174 | cpus = dl_bw_cpus(dest_cpu); | |
5175 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5176 | if (overflow) | |
5177 | ret = -EBUSY; | |
5178 | else { | |
5179 | /* | |
5180 | * We reserve space for this task in the destination | |
5181 | * root_domain, as we can't fail after this point. | |
5182 | * We will free resources in the source root_domain | |
5183 | * later on (see set_cpus_allowed_dl()). | |
5184 | */ | |
5185 | __dl_add(dl_b, p->dl.dl_bw); | |
5186 | } | |
5187 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5188 | rcu_read_unlock_sched(); |
7f51412a JL |
5189 | |
5190 | } | |
5191 | #endif | |
5192 | out: | |
5193 | return ret; | |
5194 | } | |
5195 | ||
1da177e4 | 5196 | #ifdef CONFIG_SMP |
1da177e4 | 5197 | |
e26fbffd TG |
5198 | static bool sched_smp_initialized __read_mostly; |
5199 | ||
e6628d5b MG |
5200 | #ifdef CONFIG_NUMA_BALANCING |
5201 | /* Migrate current task p to target_cpu */ | |
5202 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5203 | { | |
5204 | struct migration_arg arg = { p, target_cpu }; | |
5205 | int curr_cpu = task_cpu(p); | |
5206 | ||
5207 | if (curr_cpu == target_cpu) | |
5208 | return 0; | |
5209 | ||
5210 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5211 | return -EINVAL; | |
5212 | ||
5213 | /* TODO: This is not properly updating schedstats */ | |
5214 | ||
286549dc | 5215 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5216 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5217 | } | |
0ec8aa00 PZ |
5218 | |
5219 | /* | |
5220 | * Requeue a task on a given node and accurately track the number of NUMA | |
5221 | * tasks on the runqueues | |
5222 | */ | |
5223 | void sched_setnuma(struct task_struct *p, int nid) | |
5224 | { | |
5225 | struct rq *rq; | |
5226 | unsigned long flags; | |
da0c1e65 | 5227 | bool queued, running; |
0ec8aa00 PZ |
5228 | |
5229 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 5230 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5231 | running = task_current(rq, p); |
5232 | ||
da0c1e65 | 5233 | if (queued) |
1de64443 | 5234 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5235 | if (running) |
f3cd1c4e | 5236 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5237 | |
5238 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5239 | |
5240 | if (running) | |
5241 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5242 | if (queued) |
1de64443 | 5243 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
0ec8aa00 PZ |
5244 | task_rq_unlock(rq, p, &flags); |
5245 | } | |
5cc389bc | 5246 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5247 | |
1da177e4 | 5248 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5249 | /* |
48c5ccae PZ |
5250 | * Ensures that the idle task is using init_mm right before its cpu goes |
5251 | * offline. | |
054b9108 | 5252 | */ |
48c5ccae | 5253 | void idle_task_exit(void) |
1da177e4 | 5254 | { |
48c5ccae | 5255 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5256 | |
48c5ccae | 5257 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5258 | |
a53efe5f | 5259 | if (mm != &init_mm) { |
48c5ccae | 5260 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
5261 | finish_arch_post_lock_switch(); |
5262 | } | |
48c5ccae | 5263 | mmdrop(mm); |
1da177e4 LT |
5264 | } |
5265 | ||
5266 | /* | |
5d180232 PZ |
5267 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5268 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5269 | * nr_active count is stable. | |
5270 | * | |
5271 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5272 | */ |
5d180232 | 5273 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5274 | { |
5d180232 PZ |
5275 | long delta = calc_load_fold_active(rq); |
5276 | if (delta) | |
5277 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5278 | } |
5279 | ||
3f1d2a31 PZ |
5280 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5281 | { | |
5282 | } | |
5283 | ||
5284 | static const struct sched_class fake_sched_class = { | |
5285 | .put_prev_task = put_prev_task_fake, | |
5286 | }; | |
5287 | ||
5288 | static struct task_struct fake_task = { | |
5289 | /* | |
5290 | * Avoid pull_{rt,dl}_task() | |
5291 | */ | |
5292 | .prio = MAX_PRIO + 1, | |
5293 | .sched_class = &fake_sched_class, | |
5294 | }; | |
5295 | ||
48f24c4d | 5296 | /* |
48c5ccae PZ |
5297 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5298 | * try_to_wake_up()->select_task_rq(). | |
5299 | * | |
5300 | * Called with rq->lock held even though we'er in stop_machine() and | |
5301 | * there's no concurrency possible, we hold the required locks anyway | |
5302 | * because of lock validation efforts. | |
1da177e4 | 5303 | */ |
5e16bbc2 | 5304 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5305 | { |
5e16bbc2 | 5306 | struct rq *rq = dead_rq; |
48c5ccae PZ |
5307 | struct task_struct *next, *stop = rq->stop; |
5308 | int dest_cpu; | |
1da177e4 LT |
5309 | |
5310 | /* | |
48c5ccae PZ |
5311 | * Fudge the rq selection such that the below task selection loop |
5312 | * doesn't get stuck on the currently eligible stop task. | |
5313 | * | |
5314 | * We're currently inside stop_machine() and the rq is either stuck | |
5315 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5316 | * either way we should never end up calling schedule() until we're | |
5317 | * done here. | |
1da177e4 | 5318 | */ |
48c5ccae | 5319 | rq->stop = NULL; |
48f24c4d | 5320 | |
77bd3970 FW |
5321 | /* |
5322 | * put_prev_task() and pick_next_task() sched | |
5323 | * class method both need to have an up-to-date | |
5324 | * value of rq->clock[_task] | |
5325 | */ | |
5326 | update_rq_clock(rq); | |
5327 | ||
5e16bbc2 | 5328 | for (;;) { |
48c5ccae PZ |
5329 | /* |
5330 | * There's this thread running, bail when that's the only | |
5331 | * remaining thread. | |
5332 | */ | |
5333 | if (rq->nr_running == 1) | |
dd41f596 | 5334 | break; |
48c5ccae | 5335 | |
cbce1a68 | 5336 | /* |
5473e0cc | 5337 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 PZ |
5338 | */ |
5339 | lockdep_pin_lock(&rq->lock); | |
3f1d2a31 | 5340 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5341 | BUG_ON(!next); |
79c53799 | 5342 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5343 | |
5473e0cc WL |
5344 | /* |
5345 | * Rules for changing task_struct::cpus_allowed are holding | |
5346 | * both pi_lock and rq->lock, such that holding either | |
5347 | * stabilizes the mask. | |
5348 | * | |
5349 | * Drop rq->lock is not quite as disastrous as it usually is | |
5350 | * because !cpu_active at this point, which means load-balance | |
5351 | * will not interfere. Also, stop-machine. | |
5352 | */ | |
5353 | lockdep_unpin_lock(&rq->lock); | |
5354 | raw_spin_unlock(&rq->lock); | |
5355 | raw_spin_lock(&next->pi_lock); | |
5356 | raw_spin_lock(&rq->lock); | |
5357 | ||
5358 | /* | |
5359 | * Since we're inside stop-machine, _nothing_ should have | |
5360 | * changed the task, WARN if weird stuff happened, because in | |
5361 | * that case the above rq->lock drop is a fail too. | |
5362 | */ | |
5363 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5364 | raw_spin_unlock(&next->pi_lock); | |
5365 | continue; | |
5366 | } | |
5367 | ||
48c5ccae | 5368 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5369 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5370 | |
5e16bbc2 PZ |
5371 | rq = __migrate_task(rq, next, dest_cpu); |
5372 | if (rq != dead_rq) { | |
5373 | raw_spin_unlock(&rq->lock); | |
5374 | rq = dead_rq; | |
5375 | raw_spin_lock(&rq->lock); | |
5376 | } | |
5473e0cc | 5377 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5378 | } |
dce48a84 | 5379 | |
48c5ccae | 5380 | rq->stop = stop; |
dce48a84 | 5381 | } |
1da177e4 LT |
5382 | #endif /* CONFIG_HOTPLUG_CPU */ |
5383 | ||
1f11eb6a GH |
5384 | static void set_rq_online(struct rq *rq) |
5385 | { | |
5386 | if (!rq->online) { | |
5387 | const struct sched_class *class; | |
5388 | ||
c6c4927b | 5389 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5390 | rq->online = 1; |
5391 | ||
5392 | for_each_class(class) { | |
5393 | if (class->rq_online) | |
5394 | class->rq_online(rq); | |
5395 | } | |
5396 | } | |
5397 | } | |
5398 | ||
5399 | static void set_rq_offline(struct rq *rq) | |
5400 | { | |
5401 | if (rq->online) { | |
5402 | const struct sched_class *class; | |
5403 | ||
5404 | for_each_class(class) { | |
5405 | if (class->rq_offline) | |
5406 | class->rq_offline(rq); | |
5407 | } | |
5408 | ||
c6c4927b | 5409 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5410 | rq->online = 0; |
5411 | } | |
5412 | } | |
5413 | ||
9cf7243d | 5414 | static void set_cpu_rq_start_time(unsigned int cpu) |
a803f026 | 5415 | { |
a803f026 | 5416 | struct rq *rq = cpu_rq(cpu); |
9cf7243d | 5417 | |
a803f026 CM |
5418 | rq->age_stamp = sched_clock_cpu(cpu); |
5419 | } | |
5420 | ||
4cb98839 PZ |
5421 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5422 | ||
3e9830dc | 5423 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5424 | |
d039ac60 | 5425 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5426 | |
d039ac60 | 5427 | static int __init sched_debug_setup(char *str) |
f6630114 | 5428 | { |
d039ac60 | 5429 | sched_debug_enabled = 1; |
f6630114 MT |
5430 | |
5431 | return 0; | |
5432 | } | |
d039ac60 PZ |
5433 | early_param("sched_debug", sched_debug_setup); |
5434 | ||
5435 | static inline bool sched_debug(void) | |
5436 | { | |
5437 | return sched_debug_enabled; | |
5438 | } | |
f6630114 | 5439 | |
7c16ec58 | 5440 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5441 | struct cpumask *groupmask) |
1da177e4 | 5442 | { |
4dcf6aff | 5443 | struct sched_group *group = sd->groups; |
1da177e4 | 5444 | |
96f874e2 | 5445 | cpumask_clear(groupmask); |
4dcf6aff IM |
5446 | |
5447 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5448 | ||
5449 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5450 | printk("does not load-balance\n"); |
4dcf6aff | 5451 | if (sd->parent) |
3df0fc5b PZ |
5452 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5453 | " has parent"); | |
4dcf6aff | 5454 | return -1; |
41c7ce9a NP |
5455 | } |
5456 | ||
333470ee TH |
5457 | printk(KERN_CONT "span %*pbl level %s\n", |
5458 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5459 | |
758b2cdc | 5460 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5461 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5462 | "CPU%d\n", cpu); | |
4dcf6aff | 5463 | } |
758b2cdc | 5464 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5465 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5466 | " CPU%d\n", cpu); | |
4dcf6aff | 5467 | } |
1da177e4 | 5468 | |
4dcf6aff | 5469 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5470 | do { |
4dcf6aff | 5471 | if (!group) { |
3df0fc5b PZ |
5472 | printk("\n"); |
5473 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5474 | break; |
5475 | } | |
5476 | ||
758b2cdc | 5477 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5478 | printk(KERN_CONT "\n"); |
5479 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5480 | break; |
5481 | } | |
1da177e4 | 5482 | |
cb83b629 PZ |
5483 | if (!(sd->flags & SD_OVERLAP) && |
5484 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5485 | printk(KERN_CONT "\n"); |
5486 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5487 | break; |
5488 | } | |
1da177e4 | 5489 | |
758b2cdc | 5490 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5491 | |
333470ee TH |
5492 | printk(KERN_CONT " %*pbl", |
5493 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5494 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5495 | printk(KERN_CONT " (cpu_capacity = %d)", |
5496 | group->sgc->capacity); | |
381512cf | 5497 | } |
1da177e4 | 5498 | |
4dcf6aff IM |
5499 | group = group->next; |
5500 | } while (group != sd->groups); | |
3df0fc5b | 5501 | printk(KERN_CONT "\n"); |
1da177e4 | 5502 | |
758b2cdc | 5503 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5504 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5505 | |
758b2cdc RR |
5506 | if (sd->parent && |
5507 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5508 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5509 | "of domain->span\n"); | |
4dcf6aff IM |
5510 | return 0; |
5511 | } | |
1da177e4 | 5512 | |
4dcf6aff IM |
5513 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5514 | { | |
5515 | int level = 0; | |
1da177e4 | 5516 | |
d039ac60 | 5517 | if (!sched_debug_enabled) |
f6630114 MT |
5518 | return; |
5519 | ||
4dcf6aff IM |
5520 | if (!sd) { |
5521 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5522 | return; | |
5523 | } | |
1da177e4 | 5524 | |
4dcf6aff IM |
5525 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5526 | ||
5527 | for (;;) { | |
4cb98839 | 5528 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5529 | break; |
1da177e4 LT |
5530 | level++; |
5531 | sd = sd->parent; | |
33859f7f | 5532 | if (!sd) |
4dcf6aff IM |
5533 | break; |
5534 | } | |
1da177e4 | 5535 | } |
6d6bc0ad | 5536 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5537 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5538 | static inline bool sched_debug(void) |
5539 | { | |
5540 | return false; | |
5541 | } | |
6d6bc0ad | 5542 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5543 | |
1a20ff27 | 5544 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5545 | { |
758b2cdc | 5546 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5547 | return 1; |
5548 | ||
5549 | /* Following flags need at least 2 groups */ | |
5550 | if (sd->flags & (SD_LOAD_BALANCE | | |
5551 | SD_BALANCE_NEWIDLE | | |
5552 | SD_BALANCE_FORK | | |
89c4710e | 5553 | SD_BALANCE_EXEC | |
5d4dfddd | 5554 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5555 | SD_SHARE_PKG_RESOURCES | |
5556 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5557 | if (sd->groups != sd->groups->next) |
5558 | return 0; | |
5559 | } | |
5560 | ||
5561 | /* Following flags don't use groups */ | |
c88d5910 | 5562 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5563 | return 0; |
5564 | ||
5565 | return 1; | |
5566 | } | |
5567 | ||
48f24c4d IM |
5568 | static int |
5569 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5570 | { |
5571 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5572 | ||
5573 | if (sd_degenerate(parent)) | |
5574 | return 1; | |
5575 | ||
758b2cdc | 5576 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5577 | return 0; |
5578 | ||
245af2c7 SS |
5579 | /* Flags needing groups don't count if only 1 group in parent */ |
5580 | if (parent->groups == parent->groups->next) { | |
5581 | pflags &= ~(SD_LOAD_BALANCE | | |
5582 | SD_BALANCE_NEWIDLE | | |
5583 | SD_BALANCE_FORK | | |
89c4710e | 5584 | SD_BALANCE_EXEC | |
5d4dfddd | 5585 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5586 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5587 | SD_PREFER_SIBLING | |
5588 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5589 | if (nr_node_ids == 1) |
5590 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5591 | } |
5592 | if (~cflags & pflags) | |
5593 | return 0; | |
5594 | ||
5595 | return 1; | |
5596 | } | |
5597 | ||
dce840a0 | 5598 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5599 | { |
dce840a0 | 5600 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5601 | |
68e74568 | 5602 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5603 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5604 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5605 | free_cpumask_var(rd->rto_mask); |
5606 | free_cpumask_var(rd->online); | |
5607 | free_cpumask_var(rd->span); | |
5608 | kfree(rd); | |
5609 | } | |
5610 | ||
57d885fe GH |
5611 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5612 | { | |
a0490fa3 | 5613 | struct root_domain *old_rd = NULL; |
57d885fe | 5614 | unsigned long flags; |
57d885fe | 5615 | |
05fa785c | 5616 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5617 | |
5618 | if (rq->rd) { | |
a0490fa3 | 5619 | old_rd = rq->rd; |
57d885fe | 5620 | |
c6c4927b | 5621 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5622 | set_rq_offline(rq); |
57d885fe | 5623 | |
c6c4927b | 5624 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5625 | |
a0490fa3 | 5626 | /* |
0515973f | 5627 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5628 | * set old_rd to NULL to skip the freeing later |
5629 | * in this function: | |
5630 | */ | |
5631 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5632 | old_rd = NULL; | |
57d885fe GH |
5633 | } |
5634 | ||
5635 | atomic_inc(&rd->refcount); | |
5636 | rq->rd = rd; | |
5637 | ||
c6c4927b | 5638 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5639 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5640 | set_rq_online(rq); |
57d885fe | 5641 | |
05fa785c | 5642 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5643 | |
5644 | if (old_rd) | |
dce840a0 | 5645 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5646 | } |
5647 | ||
68c38fc3 | 5648 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5649 | { |
5650 | memset(rd, 0, sizeof(*rd)); | |
5651 | ||
8295c699 | 5652 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5653 | goto out; |
8295c699 | 5654 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5655 | goto free_span; |
8295c699 | 5656 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5657 | goto free_online; |
8295c699 | 5658 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5659 | goto free_dlo_mask; |
6e0534f2 | 5660 | |
332ac17e | 5661 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5662 | if (cpudl_init(&rd->cpudl) != 0) |
5663 | goto free_dlo_mask; | |
332ac17e | 5664 | |
68c38fc3 | 5665 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5666 | goto free_rto_mask; |
c6c4927b | 5667 | return 0; |
6e0534f2 | 5668 | |
68e74568 RR |
5669 | free_rto_mask: |
5670 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5671 | free_dlo_mask: |
5672 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5673 | free_online: |
5674 | free_cpumask_var(rd->online); | |
5675 | free_span: | |
5676 | free_cpumask_var(rd->span); | |
0c910d28 | 5677 | out: |
c6c4927b | 5678 | return -ENOMEM; |
57d885fe GH |
5679 | } |
5680 | ||
029632fb PZ |
5681 | /* |
5682 | * By default the system creates a single root-domain with all cpus as | |
5683 | * members (mimicking the global state we have today). | |
5684 | */ | |
5685 | struct root_domain def_root_domain; | |
5686 | ||
57d885fe GH |
5687 | static void init_defrootdomain(void) |
5688 | { | |
68c38fc3 | 5689 | init_rootdomain(&def_root_domain); |
c6c4927b | 5690 | |
57d885fe GH |
5691 | atomic_set(&def_root_domain.refcount, 1); |
5692 | } | |
5693 | ||
dc938520 | 5694 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5695 | { |
5696 | struct root_domain *rd; | |
5697 | ||
5698 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5699 | if (!rd) | |
5700 | return NULL; | |
5701 | ||
68c38fc3 | 5702 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5703 | kfree(rd); |
5704 | return NULL; | |
5705 | } | |
57d885fe GH |
5706 | |
5707 | return rd; | |
5708 | } | |
5709 | ||
63b2ca30 | 5710 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5711 | { |
5712 | struct sched_group *tmp, *first; | |
5713 | ||
5714 | if (!sg) | |
5715 | return; | |
5716 | ||
5717 | first = sg; | |
5718 | do { | |
5719 | tmp = sg->next; | |
5720 | ||
63b2ca30 NP |
5721 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5722 | kfree(sg->sgc); | |
e3589f6c PZ |
5723 | |
5724 | kfree(sg); | |
5725 | sg = tmp; | |
5726 | } while (sg != first); | |
5727 | } | |
5728 | ||
dce840a0 PZ |
5729 | static void free_sched_domain(struct rcu_head *rcu) |
5730 | { | |
5731 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5732 | |
5733 | /* | |
5734 | * If its an overlapping domain it has private groups, iterate and | |
5735 | * nuke them all. | |
5736 | */ | |
5737 | if (sd->flags & SD_OVERLAP) { | |
5738 | free_sched_groups(sd->groups, 1); | |
5739 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5740 | kfree(sd->groups->sgc); |
dce840a0 | 5741 | kfree(sd->groups); |
9c3f75cb | 5742 | } |
dce840a0 PZ |
5743 | kfree(sd); |
5744 | } | |
5745 | ||
5746 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5747 | { | |
5748 | call_rcu(&sd->rcu, free_sched_domain); | |
5749 | } | |
5750 | ||
5751 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5752 | { | |
5753 | for (; sd; sd = sd->parent) | |
5754 | destroy_sched_domain(sd, cpu); | |
5755 | } | |
5756 | ||
518cd623 PZ |
5757 | /* |
5758 | * Keep a special pointer to the highest sched_domain that has | |
5759 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5760 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5761 | * | |
5762 | * Also keep a unique ID per domain (we use the first cpu number in | |
5763 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5764 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5765 | */ |
5766 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5767 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5768 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5769 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5770 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5771 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5772 | |
5773 | static void update_top_cache_domain(int cpu) | |
5774 | { | |
5775 | struct sched_domain *sd; | |
5d4cf996 | 5776 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5777 | int id = cpu; |
7d9ffa89 | 5778 | int size = 1; |
518cd623 PZ |
5779 | |
5780 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5781 | if (sd) { |
518cd623 | 5782 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5783 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5784 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5785 | } |
5d4cf996 | 5786 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5787 | |
5788 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5789 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5790 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5791 | |
5792 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5793 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5794 | |
5795 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5796 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5797 | } |
5798 | ||
1da177e4 | 5799 | /* |
0eab9146 | 5800 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5801 | * hold the hotplug lock. |
5802 | */ | |
0eab9146 IM |
5803 | static void |
5804 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5805 | { |
70b97a7f | 5806 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5807 | struct sched_domain *tmp; |
5808 | ||
5809 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5810 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5811 | struct sched_domain *parent = tmp->parent; |
5812 | if (!parent) | |
5813 | break; | |
f29c9b1c | 5814 | |
1a848870 | 5815 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5816 | tmp->parent = parent->parent; |
1a848870 SS |
5817 | if (parent->parent) |
5818 | parent->parent->child = tmp; | |
10866e62 PZ |
5819 | /* |
5820 | * Transfer SD_PREFER_SIBLING down in case of a | |
5821 | * degenerate parent; the spans match for this | |
5822 | * so the property transfers. | |
5823 | */ | |
5824 | if (parent->flags & SD_PREFER_SIBLING) | |
5825 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5826 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5827 | } else |
5828 | tmp = tmp->parent; | |
245af2c7 SS |
5829 | } |
5830 | ||
1a848870 | 5831 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5832 | tmp = sd; |
245af2c7 | 5833 | sd = sd->parent; |
dce840a0 | 5834 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5835 | if (sd) |
5836 | sd->child = NULL; | |
5837 | } | |
1da177e4 | 5838 | |
4cb98839 | 5839 | sched_domain_debug(sd, cpu); |
1da177e4 | 5840 | |
57d885fe | 5841 | rq_attach_root(rq, rd); |
dce840a0 | 5842 | tmp = rq->sd; |
674311d5 | 5843 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5844 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5845 | |
5846 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5847 | } |
5848 | ||
1da177e4 LT |
5849 | /* Setup the mask of cpus configured for isolated domains */ |
5850 | static int __init isolated_cpu_setup(char *str) | |
5851 | { | |
a6e4491c PB |
5852 | int ret; |
5853 | ||
bdddd296 | 5854 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
5855 | ret = cpulist_parse(str, cpu_isolated_map); |
5856 | if (ret) { | |
5857 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
5858 | return 0; | |
5859 | } | |
1da177e4 LT |
5860 | return 1; |
5861 | } | |
8927f494 | 5862 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5863 | |
49a02c51 | 5864 | struct s_data { |
21d42ccf | 5865 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5866 | struct root_domain *rd; |
5867 | }; | |
5868 | ||
2109b99e | 5869 | enum s_alloc { |
2109b99e | 5870 | sa_rootdomain, |
21d42ccf | 5871 | sa_sd, |
dce840a0 | 5872 | sa_sd_storage, |
2109b99e AH |
5873 | sa_none, |
5874 | }; | |
5875 | ||
c1174876 PZ |
5876 | /* |
5877 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5878 | * domain traversal. | |
5879 | * | |
5880 | * Asymmetric node setups can result in situations where the domain tree is of | |
5881 | * unequal depth, make sure to skip domains that already cover the entire | |
5882 | * range. | |
5883 | * | |
5884 | * In that case build_sched_domains() will have terminated the iteration early | |
5885 | * and our sibling sd spans will be empty. Domains should always include the | |
5886 | * cpu they're built on, so check that. | |
5887 | * | |
5888 | */ | |
5889 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5890 | { | |
5891 | const struct cpumask *span = sched_domain_span(sd); | |
5892 | struct sd_data *sdd = sd->private; | |
5893 | struct sched_domain *sibling; | |
5894 | int i; | |
5895 | ||
5896 | for_each_cpu(i, span) { | |
5897 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5898 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5899 | continue; | |
5900 | ||
5901 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5902 | } | |
5903 | } | |
5904 | ||
5905 | /* | |
5906 | * Return the canonical balance cpu for this group, this is the first cpu | |
5907 | * of this group that's also in the iteration mask. | |
5908 | */ | |
5909 | int group_balance_cpu(struct sched_group *sg) | |
5910 | { | |
5911 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5912 | } | |
5913 | ||
e3589f6c PZ |
5914 | static int |
5915 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5916 | { | |
5917 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5918 | const struct cpumask *span = sched_domain_span(sd); | |
5919 | struct cpumask *covered = sched_domains_tmpmask; | |
5920 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5921 | struct sched_domain *sibling; |
e3589f6c PZ |
5922 | int i; |
5923 | ||
5924 | cpumask_clear(covered); | |
5925 | ||
5926 | for_each_cpu(i, span) { | |
5927 | struct cpumask *sg_span; | |
5928 | ||
5929 | if (cpumask_test_cpu(i, covered)) | |
5930 | continue; | |
5931 | ||
aaecac4a | 5932 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5933 | |
5934 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5935 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5936 | continue; |
5937 | ||
e3589f6c | 5938 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5939 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5940 | |
5941 | if (!sg) | |
5942 | goto fail; | |
5943 | ||
5944 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5945 | if (sibling->child) |
5946 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5947 | else | |
e3589f6c PZ |
5948 | cpumask_set_cpu(i, sg_span); |
5949 | ||
5950 | cpumask_or(covered, covered, sg_span); | |
5951 | ||
63b2ca30 NP |
5952 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5953 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5954 | build_group_mask(sd, sg); |
5955 | ||
c3decf0d | 5956 | /* |
63b2ca30 | 5957 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5958 | * domains and no possible iteration will get us here, we won't |
5959 | * die on a /0 trap. | |
5960 | */ | |
ca8ce3d0 | 5961 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 5962 | |
c1174876 PZ |
5963 | /* |
5964 | * Make sure the first group of this domain contains the | |
5965 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5966 | * breaks. See update_sg_lb_stats(). | |
5967 | */ | |
74a5ce20 | 5968 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5969 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5970 | groups = sg; |
5971 | ||
5972 | if (!first) | |
5973 | first = sg; | |
5974 | if (last) | |
5975 | last->next = sg; | |
5976 | last = sg; | |
5977 | last->next = first; | |
5978 | } | |
5979 | sd->groups = groups; | |
5980 | ||
5981 | return 0; | |
5982 | ||
5983 | fail: | |
5984 | free_sched_groups(first, 0); | |
5985 | ||
5986 | return -ENOMEM; | |
5987 | } | |
5988 | ||
dce840a0 | 5989 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5990 | { |
dce840a0 PZ |
5991 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5992 | struct sched_domain *child = sd->child; | |
1da177e4 | 5993 | |
dce840a0 PZ |
5994 | if (child) |
5995 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5996 | |
9c3f75cb | 5997 | if (sg) { |
dce840a0 | 5998 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5999 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6000 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6001 | } |
dce840a0 PZ |
6002 | |
6003 | return cpu; | |
1e9f28fa | 6004 | } |
1e9f28fa | 6005 | |
01a08546 | 6006 | /* |
dce840a0 PZ |
6007 | * build_sched_groups will build a circular linked list of the groups |
6008 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6009 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6010 | * |
6011 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6012 | */ |
e3589f6c PZ |
6013 | static int |
6014 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6015 | { |
dce840a0 PZ |
6016 | struct sched_group *first = NULL, *last = NULL; |
6017 | struct sd_data *sdd = sd->private; | |
6018 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6019 | struct cpumask *covered; |
dce840a0 | 6020 | int i; |
9c1cfda2 | 6021 | |
e3589f6c PZ |
6022 | get_group(cpu, sdd, &sd->groups); |
6023 | atomic_inc(&sd->groups->ref); | |
6024 | ||
0936629f | 6025 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6026 | return 0; |
6027 | ||
f96225fd PZ |
6028 | lockdep_assert_held(&sched_domains_mutex); |
6029 | covered = sched_domains_tmpmask; | |
6030 | ||
dce840a0 | 6031 | cpumask_clear(covered); |
6711cab4 | 6032 | |
dce840a0 PZ |
6033 | for_each_cpu(i, span) { |
6034 | struct sched_group *sg; | |
cd08e923 | 6035 | int group, j; |
6711cab4 | 6036 | |
dce840a0 PZ |
6037 | if (cpumask_test_cpu(i, covered)) |
6038 | continue; | |
6711cab4 | 6039 | |
cd08e923 | 6040 | group = get_group(i, sdd, &sg); |
c1174876 | 6041 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6042 | |
dce840a0 PZ |
6043 | for_each_cpu(j, span) { |
6044 | if (get_group(j, sdd, NULL) != group) | |
6045 | continue; | |
0601a88d | 6046 | |
dce840a0 PZ |
6047 | cpumask_set_cpu(j, covered); |
6048 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6049 | } | |
0601a88d | 6050 | |
dce840a0 PZ |
6051 | if (!first) |
6052 | first = sg; | |
6053 | if (last) | |
6054 | last->next = sg; | |
6055 | last = sg; | |
6056 | } | |
6057 | last->next = first; | |
e3589f6c PZ |
6058 | |
6059 | return 0; | |
0601a88d | 6060 | } |
51888ca2 | 6061 | |
89c4710e | 6062 | /* |
63b2ca30 | 6063 | * Initialize sched groups cpu_capacity. |
89c4710e | 6064 | * |
63b2ca30 | 6065 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6066 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6067 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6068 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6069 | * group having more cpu_capacity will pickup more load compared to the | |
6070 | * group having less cpu_capacity. | |
89c4710e | 6071 | */ |
63b2ca30 | 6072 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6073 | { |
e3589f6c | 6074 | struct sched_group *sg = sd->groups; |
89c4710e | 6075 | |
94c95ba6 | 6076 | WARN_ON(!sg); |
e3589f6c PZ |
6077 | |
6078 | do { | |
6079 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6080 | sg = sg->next; | |
6081 | } while (sg != sd->groups); | |
89c4710e | 6082 | |
c1174876 | 6083 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6084 | return; |
aae6d3dd | 6085 | |
63b2ca30 NP |
6086 | update_group_capacity(sd, cpu); |
6087 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6088 | } |
6089 | ||
7c16ec58 MT |
6090 | /* |
6091 | * Initializers for schedule domains | |
6092 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6093 | */ | |
6094 | ||
1d3504fc | 6095 | static int default_relax_domain_level = -1; |
60495e77 | 6096 | int sched_domain_level_max; |
1d3504fc HS |
6097 | |
6098 | static int __init setup_relax_domain_level(char *str) | |
6099 | { | |
a841f8ce DS |
6100 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6101 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6102 | |
1d3504fc HS |
6103 | return 1; |
6104 | } | |
6105 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6106 | ||
6107 | static void set_domain_attribute(struct sched_domain *sd, | |
6108 | struct sched_domain_attr *attr) | |
6109 | { | |
6110 | int request; | |
6111 | ||
6112 | if (!attr || attr->relax_domain_level < 0) { | |
6113 | if (default_relax_domain_level < 0) | |
6114 | return; | |
6115 | else | |
6116 | request = default_relax_domain_level; | |
6117 | } else | |
6118 | request = attr->relax_domain_level; | |
6119 | if (request < sd->level) { | |
6120 | /* turn off idle balance on this domain */ | |
c88d5910 | 6121 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6122 | } else { |
6123 | /* turn on idle balance on this domain */ | |
c88d5910 | 6124 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6125 | } |
6126 | } | |
6127 | ||
54ab4ff4 PZ |
6128 | static void __sdt_free(const struct cpumask *cpu_map); |
6129 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6130 | ||
2109b99e AH |
6131 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6132 | const struct cpumask *cpu_map) | |
6133 | { | |
6134 | switch (what) { | |
2109b99e | 6135 | case sa_rootdomain: |
822ff793 PZ |
6136 | if (!atomic_read(&d->rd->refcount)) |
6137 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6138 | case sa_sd: |
6139 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6140 | case sa_sd_storage: |
54ab4ff4 | 6141 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6142 | case sa_none: |
6143 | break; | |
6144 | } | |
6145 | } | |
3404c8d9 | 6146 | |
2109b99e AH |
6147 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6148 | const struct cpumask *cpu_map) | |
6149 | { | |
dce840a0 PZ |
6150 | memset(d, 0, sizeof(*d)); |
6151 | ||
54ab4ff4 PZ |
6152 | if (__sdt_alloc(cpu_map)) |
6153 | return sa_sd_storage; | |
dce840a0 PZ |
6154 | d->sd = alloc_percpu(struct sched_domain *); |
6155 | if (!d->sd) | |
6156 | return sa_sd_storage; | |
2109b99e | 6157 | d->rd = alloc_rootdomain(); |
dce840a0 | 6158 | if (!d->rd) |
21d42ccf | 6159 | return sa_sd; |
2109b99e AH |
6160 | return sa_rootdomain; |
6161 | } | |
57d885fe | 6162 | |
dce840a0 PZ |
6163 | /* |
6164 | * NULL the sd_data elements we've used to build the sched_domain and | |
6165 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6166 | * will not free the data we're using. | |
6167 | */ | |
6168 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6169 | { | |
6170 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6171 | |
6172 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6173 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6174 | ||
e3589f6c | 6175 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6176 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6177 | |
63b2ca30 NP |
6178 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6179 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6180 | } |
6181 | ||
cb83b629 | 6182 | #ifdef CONFIG_NUMA |
cb83b629 | 6183 | static int sched_domains_numa_levels; |
e3fe70b1 | 6184 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6185 | static int *sched_domains_numa_distance; |
9942f79b | 6186 | int sched_max_numa_distance; |
cb83b629 PZ |
6187 | static struct cpumask ***sched_domains_numa_masks; |
6188 | static int sched_domains_curr_level; | |
143e1e28 | 6189 | #endif |
cb83b629 | 6190 | |
143e1e28 VG |
6191 | /* |
6192 | * SD_flags allowed in topology descriptions. | |
6193 | * | |
5d4dfddd | 6194 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6195 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6196 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6197 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6198 | * |
6199 | * Odd one out: | |
6200 | * SD_ASYM_PACKING - describes SMT quirks | |
6201 | */ | |
6202 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6203 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6204 | SD_SHARE_PKG_RESOURCES | \ |
6205 | SD_NUMA | \ | |
d77b3ed5 VG |
6206 | SD_ASYM_PACKING | \ |
6207 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6208 | |
6209 | static struct sched_domain * | |
143e1e28 | 6210 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6211 | { |
6212 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6213 | int sd_weight, sd_flags = 0; |
6214 | ||
6215 | #ifdef CONFIG_NUMA | |
6216 | /* | |
6217 | * Ugly hack to pass state to sd_numa_mask()... | |
6218 | */ | |
6219 | sched_domains_curr_level = tl->numa_level; | |
6220 | #endif | |
6221 | ||
6222 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6223 | ||
6224 | if (tl->sd_flags) | |
6225 | sd_flags = (*tl->sd_flags)(); | |
6226 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6227 | "wrong sd_flags in topology description\n")) | |
6228 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6229 | |
6230 | *sd = (struct sched_domain){ | |
6231 | .min_interval = sd_weight, | |
6232 | .max_interval = 2*sd_weight, | |
6233 | .busy_factor = 32, | |
870a0bb5 | 6234 | .imbalance_pct = 125, |
143e1e28 VG |
6235 | |
6236 | .cache_nice_tries = 0, | |
6237 | .busy_idx = 0, | |
6238 | .idle_idx = 0, | |
cb83b629 PZ |
6239 | .newidle_idx = 0, |
6240 | .wake_idx = 0, | |
6241 | .forkexec_idx = 0, | |
6242 | ||
6243 | .flags = 1*SD_LOAD_BALANCE | |
6244 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6245 | | 1*SD_BALANCE_EXEC |
6246 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6247 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6248 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6249 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6250 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6251 | | 0*SD_SERIALIZE |
cb83b629 | 6252 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6253 | | 0*SD_NUMA |
6254 | | sd_flags | |
cb83b629 | 6255 | , |
143e1e28 | 6256 | |
cb83b629 PZ |
6257 | .last_balance = jiffies, |
6258 | .balance_interval = sd_weight, | |
143e1e28 | 6259 | .smt_gain = 0, |
2b4cfe64 JL |
6260 | .max_newidle_lb_cost = 0, |
6261 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6262 | #ifdef CONFIG_SCHED_DEBUG |
6263 | .name = tl->name, | |
6264 | #endif | |
cb83b629 | 6265 | }; |
cb83b629 PZ |
6266 | |
6267 | /* | |
143e1e28 | 6268 | * Convert topological properties into behaviour. |
cb83b629 | 6269 | */ |
143e1e28 | 6270 | |
5d4dfddd | 6271 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6272 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6273 | sd->imbalance_pct = 110; |
6274 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6275 | |
6276 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6277 | sd->imbalance_pct = 117; | |
6278 | sd->cache_nice_tries = 1; | |
6279 | sd->busy_idx = 2; | |
6280 | ||
6281 | #ifdef CONFIG_NUMA | |
6282 | } else if (sd->flags & SD_NUMA) { | |
6283 | sd->cache_nice_tries = 2; | |
6284 | sd->busy_idx = 3; | |
6285 | sd->idle_idx = 2; | |
6286 | ||
6287 | sd->flags |= SD_SERIALIZE; | |
6288 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6289 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6290 | SD_BALANCE_FORK | | |
6291 | SD_WAKE_AFFINE); | |
6292 | } | |
6293 | ||
6294 | #endif | |
6295 | } else { | |
6296 | sd->flags |= SD_PREFER_SIBLING; | |
6297 | sd->cache_nice_tries = 1; | |
6298 | sd->busy_idx = 2; | |
6299 | sd->idle_idx = 1; | |
6300 | } | |
6301 | ||
6302 | sd->private = &tl->data; | |
cb83b629 PZ |
6303 | |
6304 | return sd; | |
6305 | } | |
6306 | ||
143e1e28 VG |
6307 | /* |
6308 | * Topology list, bottom-up. | |
6309 | */ | |
6310 | static struct sched_domain_topology_level default_topology[] = { | |
6311 | #ifdef CONFIG_SCHED_SMT | |
6312 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6313 | #endif | |
6314 | #ifdef CONFIG_SCHED_MC | |
6315 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6316 | #endif |
6317 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6318 | { NULL, }, | |
6319 | }; | |
6320 | ||
c6e1e7b5 JG |
6321 | static struct sched_domain_topology_level *sched_domain_topology = |
6322 | default_topology; | |
143e1e28 VG |
6323 | |
6324 | #define for_each_sd_topology(tl) \ | |
6325 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6326 | ||
6327 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6328 | { | |
6329 | sched_domain_topology = tl; | |
6330 | } | |
6331 | ||
6332 | #ifdef CONFIG_NUMA | |
6333 | ||
cb83b629 PZ |
6334 | static const struct cpumask *sd_numa_mask(int cpu) |
6335 | { | |
6336 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6337 | } | |
6338 | ||
d039ac60 PZ |
6339 | static void sched_numa_warn(const char *str) |
6340 | { | |
6341 | static int done = false; | |
6342 | int i,j; | |
6343 | ||
6344 | if (done) | |
6345 | return; | |
6346 | ||
6347 | done = true; | |
6348 | ||
6349 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6350 | ||
6351 | for (i = 0; i < nr_node_ids; i++) { | |
6352 | printk(KERN_WARNING " "); | |
6353 | for (j = 0; j < nr_node_ids; j++) | |
6354 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6355 | printk(KERN_CONT "\n"); | |
6356 | } | |
6357 | printk(KERN_WARNING "\n"); | |
6358 | } | |
6359 | ||
9942f79b | 6360 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6361 | { |
6362 | int i; | |
6363 | ||
6364 | if (distance == node_distance(0, 0)) | |
6365 | return true; | |
6366 | ||
6367 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6368 | if (sched_domains_numa_distance[i] == distance) | |
6369 | return true; | |
6370 | } | |
6371 | ||
6372 | return false; | |
6373 | } | |
6374 | ||
e3fe70b1 RR |
6375 | /* |
6376 | * A system can have three types of NUMA topology: | |
6377 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6378 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6379 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6380 | * | |
6381 | * The difference between a glueless mesh topology and a backplane | |
6382 | * topology lies in whether communication between not directly | |
6383 | * connected nodes goes through intermediary nodes (where programs | |
6384 | * could run), or through backplane controllers. This affects | |
6385 | * placement of programs. | |
6386 | * | |
6387 | * The type of topology can be discerned with the following tests: | |
6388 | * - If the maximum distance between any nodes is 1 hop, the system | |
6389 | * is directly connected. | |
6390 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6391 | * there is an intermediary node C, which is < N hops away from both | |
6392 | * nodes A and B, the system is a glueless mesh. | |
6393 | */ | |
6394 | static void init_numa_topology_type(void) | |
6395 | { | |
6396 | int a, b, c, n; | |
6397 | ||
6398 | n = sched_max_numa_distance; | |
6399 | ||
e237882b | 6400 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6401 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6402 | return; |
6403 | } | |
e3fe70b1 RR |
6404 | |
6405 | for_each_online_node(a) { | |
6406 | for_each_online_node(b) { | |
6407 | /* Find two nodes furthest removed from each other. */ | |
6408 | if (node_distance(a, b) < n) | |
6409 | continue; | |
6410 | ||
6411 | /* Is there an intermediary node between a and b? */ | |
6412 | for_each_online_node(c) { | |
6413 | if (node_distance(a, c) < n && | |
6414 | node_distance(b, c) < n) { | |
6415 | sched_numa_topology_type = | |
6416 | NUMA_GLUELESS_MESH; | |
6417 | return; | |
6418 | } | |
6419 | } | |
6420 | ||
6421 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6422 | return; | |
6423 | } | |
6424 | } | |
6425 | } | |
6426 | ||
cb83b629 PZ |
6427 | static void sched_init_numa(void) |
6428 | { | |
6429 | int next_distance, curr_distance = node_distance(0, 0); | |
6430 | struct sched_domain_topology_level *tl; | |
6431 | int level = 0; | |
6432 | int i, j, k; | |
6433 | ||
cb83b629 PZ |
6434 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6435 | if (!sched_domains_numa_distance) | |
6436 | return; | |
6437 | ||
6438 | /* | |
6439 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6440 | * unique distances in the node_distance() table. | |
6441 | * | |
6442 | * Assumes node_distance(0,j) includes all distances in | |
6443 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6444 | */ |
6445 | next_distance = curr_distance; | |
6446 | for (i = 0; i < nr_node_ids; i++) { | |
6447 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6448 | for (k = 0; k < nr_node_ids; k++) { |
6449 | int distance = node_distance(i, k); | |
6450 | ||
6451 | if (distance > curr_distance && | |
6452 | (distance < next_distance || | |
6453 | next_distance == curr_distance)) | |
6454 | next_distance = distance; | |
6455 | ||
6456 | /* | |
6457 | * While not a strong assumption it would be nice to know | |
6458 | * about cases where if node A is connected to B, B is not | |
6459 | * equally connected to A. | |
6460 | */ | |
6461 | if (sched_debug() && node_distance(k, i) != distance) | |
6462 | sched_numa_warn("Node-distance not symmetric"); | |
6463 | ||
6464 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6465 | sched_numa_warn("Node-0 not representative"); | |
6466 | } | |
6467 | if (next_distance != curr_distance) { | |
6468 | sched_domains_numa_distance[level++] = next_distance; | |
6469 | sched_domains_numa_levels = level; | |
6470 | curr_distance = next_distance; | |
6471 | } else break; | |
cb83b629 | 6472 | } |
d039ac60 PZ |
6473 | |
6474 | /* | |
6475 | * In case of sched_debug() we verify the above assumption. | |
6476 | */ | |
6477 | if (!sched_debug()) | |
6478 | break; | |
cb83b629 | 6479 | } |
c123588b AR |
6480 | |
6481 | if (!level) | |
6482 | return; | |
6483 | ||
cb83b629 PZ |
6484 | /* |
6485 | * 'level' contains the number of unique distances, excluding the | |
6486 | * identity distance node_distance(i,i). | |
6487 | * | |
28b4a521 | 6488 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6489 | * numbers. |
6490 | */ | |
6491 | ||
5f7865f3 TC |
6492 | /* |
6493 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6494 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6495 | * the array will contain less then 'level' members. This could be | |
6496 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6497 | * in other functions. | |
6498 | * | |
6499 | * We reset it to 'level' at the end of this function. | |
6500 | */ | |
6501 | sched_domains_numa_levels = 0; | |
6502 | ||
cb83b629 PZ |
6503 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6504 | if (!sched_domains_numa_masks) | |
6505 | return; | |
6506 | ||
6507 | /* | |
6508 | * Now for each level, construct a mask per node which contains all | |
6509 | * cpus of nodes that are that many hops away from us. | |
6510 | */ | |
6511 | for (i = 0; i < level; i++) { | |
6512 | sched_domains_numa_masks[i] = | |
6513 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6514 | if (!sched_domains_numa_masks[i]) | |
6515 | return; | |
6516 | ||
6517 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6518 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6519 | if (!mask) |
6520 | return; | |
6521 | ||
6522 | sched_domains_numa_masks[i][j] = mask; | |
6523 | ||
9c03ee14 | 6524 | for_each_node(k) { |
dd7d8634 | 6525 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6526 | continue; |
6527 | ||
6528 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6529 | } | |
6530 | } | |
6531 | } | |
6532 | ||
143e1e28 VG |
6533 | /* Compute default topology size */ |
6534 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6535 | ||
c515db8c | 6536 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6537 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6538 | if (!tl) | |
6539 | return; | |
6540 | ||
6541 | /* | |
6542 | * Copy the default topology bits.. | |
6543 | */ | |
143e1e28 VG |
6544 | for (i = 0; sched_domain_topology[i].mask; i++) |
6545 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6546 | |
6547 | /* | |
6548 | * .. and append 'j' levels of NUMA goodness. | |
6549 | */ | |
6550 | for (j = 0; j < level; i++, j++) { | |
6551 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6552 | .mask = sd_numa_mask, |
143e1e28 | 6553 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6554 | .flags = SDTL_OVERLAP, |
6555 | .numa_level = j, | |
143e1e28 | 6556 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6557 | }; |
6558 | } | |
6559 | ||
6560 | sched_domain_topology = tl; | |
5f7865f3 TC |
6561 | |
6562 | sched_domains_numa_levels = level; | |
9942f79b | 6563 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6564 | |
6565 | init_numa_topology_type(); | |
cb83b629 | 6566 | } |
301a5cba | 6567 | |
135fb3e1 | 6568 | static void sched_domains_numa_masks_set(unsigned int cpu) |
301a5cba | 6569 | { |
301a5cba | 6570 | int node = cpu_to_node(cpu); |
135fb3e1 TG |
6571 | int i, j; |
6572 | ||
301a5cba TC |
6573 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6574 | for (j = 0; j < nr_node_ids; j++) { | |
6575 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6576 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6577 | } | |
6578 | } | |
6579 | } | |
6580 | ||
135fb3e1 | 6581 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
301a5cba TC |
6582 | { |
6583 | int i, j; | |
135fb3e1 | 6584 | |
301a5cba TC |
6585 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6586 | for (j = 0; j < nr_node_ids; j++) | |
6587 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6588 | } | |
6589 | } | |
6590 | ||
cb83b629 | 6591 | #else |
135fb3e1 TG |
6592 | static inline void sched_init_numa(void) { } |
6593 | static void sched_domains_numa_masks_set(unsigned int cpu) { } | |
6594 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
cb83b629 PZ |
6595 | #endif /* CONFIG_NUMA */ |
6596 | ||
54ab4ff4 PZ |
6597 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6598 | { | |
6599 | struct sched_domain_topology_level *tl; | |
6600 | int j; | |
6601 | ||
27723a68 | 6602 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6603 | struct sd_data *sdd = &tl->data; |
6604 | ||
6605 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6606 | if (!sdd->sd) | |
6607 | return -ENOMEM; | |
6608 | ||
6609 | sdd->sg = alloc_percpu(struct sched_group *); | |
6610 | if (!sdd->sg) | |
6611 | return -ENOMEM; | |
6612 | ||
63b2ca30 NP |
6613 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6614 | if (!sdd->sgc) | |
9c3f75cb PZ |
6615 | return -ENOMEM; |
6616 | ||
54ab4ff4 PZ |
6617 | for_each_cpu(j, cpu_map) { |
6618 | struct sched_domain *sd; | |
6619 | struct sched_group *sg; | |
63b2ca30 | 6620 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6621 | |
5cc389bc | 6622 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6623 | GFP_KERNEL, cpu_to_node(j)); |
6624 | if (!sd) | |
6625 | return -ENOMEM; | |
6626 | ||
6627 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6628 | ||
6629 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6630 | GFP_KERNEL, cpu_to_node(j)); | |
6631 | if (!sg) | |
6632 | return -ENOMEM; | |
6633 | ||
30b4e9eb IM |
6634 | sg->next = sg; |
6635 | ||
54ab4ff4 | 6636 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6637 | |
63b2ca30 | 6638 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6639 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6640 | if (!sgc) |
9c3f75cb PZ |
6641 | return -ENOMEM; |
6642 | ||
63b2ca30 | 6643 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6644 | } |
6645 | } | |
6646 | ||
6647 | return 0; | |
6648 | } | |
6649 | ||
6650 | static void __sdt_free(const struct cpumask *cpu_map) | |
6651 | { | |
6652 | struct sched_domain_topology_level *tl; | |
6653 | int j; | |
6654 | ||
27723a68 | 6655 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6656 | struct sd_data *sdd = &tl->data; |
6657 | ||
6658 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6659 | struct sched_domain *sd; |
6660 | ||
6661 | if (sdd->sd) { | |
6662 | sd = *per_cpu_ptr(sdd->sd, j); | |
6663 | if (sd && (sd->flags & SD_OVERLAP)) | |
6664 | free_sched_groups(sd->groups, 0); | |
6665 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6666 | } | |
6667 | ||
6668 | if (sdd->sg) | |
6669 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6670 | if (sdd->sgc) |
6671 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6672 | } |
6673 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6674 | sdd->sd = NULL; |
54ab4ff4 | 6675 | free_percpu(sdd->sg); |
fb2cf2c6 | 6676 | sdd->sg = NULL; |
63b2ca30 NP |
6677 | free_percpu(sdd->sgc); |
6678 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6679 | } |
6680 | } | |
6681 | ||
2c402dc3 | 6682 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6683 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6684 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6685 | { |
143e1e28 | 6686 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6687 | if (!sd) |
d069b916 | 6688 | return child; |
2c402dc3 | 6689 | |
2c402dc3 | 6690 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6691 | if (child) { |
6692 | sd->level = child->level + 1; | |
6693 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6694 | child->parent = sd; |
c75e0128 | 6695 | sd->child = child; |
6ae72dff PZ |
6696 | |
6697 | if (!cpumask_subset(sched_domain_span(child), | |
6698 | sched_domain_span(sd))) { | |
6699 | pr_err("BUG: arch topology borken\n"); | |
6700 | #ifdef CONFIG_SCHED_DEBUG | |
6701 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6702 | child->name, sd->name); | |
6703 | #endif | |
6704 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6705 | cpumask_or(sched_domain_span(sd), | |
6706 | sched_domain_span(sd), | |
6707 | sched_domain_span(child)); | |
6708 | } | |
6709 | ||
60495e77 | 6710 | } |
a841f8ce | 6711 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6712 | |
6713 | return sd; | |
6714 | } | |
6715 | ||
2109b99e AH |
6716 | /* |
6717 | * Build sched domains for a given set of cpus and attach the sched domains | |
6718 | * to the individual cpus | |
6719 | */ | |
dce840a0 PZ |
6720 | static int build_sched_domains(const struct cpumask *cpu_map, |
6721 | struct sched_domain_attr *attr) | |
2109b99e | 6722 | { |
1c632169 | 6723 | enum s_alloc alloc_state; |
dce840a0 | 6724 | struct sched_domain *sd; |
2109b99e | 6725 | struct s_data d; |
822ff793 | 6726 | int i, ret = -ENOMEM; |
9c1cfda2 | 6727 | |
2109b99e AH |
6728 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6729 | if (alloc_state != sa_rootdomain) | |
6730 | goto error; | |
9c1cfda2 | 6731 | |
dce840a0 | 6732 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6733 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6734 | struct sched_domain_topology_level *tl; |
6735 | ||
3bd65a80 | 6736 | sd = NULL; |
27723a68 | 6737 | for_each_sd_topology(tl) { |
4a850cbe | 6738 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6739 | if (tl == sched_domain_topology) |
6740 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6741 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6742 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6743 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6744 | break; | |
e3589f6c | 6745 | } |
dce840a0 PZ |
6746 | } |
6747 | ||
6748 | /* Build the groups for the domains */ | |
6749 | for_each_cpu(i, cpu_map) { | |
6750 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6751 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6752 | if (sd->flags & SD_OVERLAP) { |
6753 | if (build_overlap_sched_groups(sd, i)) | |
6754 | goto error; | |
6755 | } else { | |
6756 | if (build_sched_groups(sd, i)) | |
6757 | goto error; | |
6758 | } | |
1cf51902 | 6759 | } |
a06dadbe | 6760 | } |
9c1cfda2 | 6761 | |
ced549fa | 6762 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6763 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6764 | if (!cpumask_test_cpu(i, cpu_map)) | |
6765 | continue; | |
9c1cfda2 | 6766 | |
dce840a0 PZ |
6767 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6768 | claim_allocations(i, sd); | |
63b2ca30 | 6769 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6770 | } |
f712c0c7 | 6771 | } |
9c1cfda2 | 6772 | |
1da177e4 | 6773 | /* Attach the domains */ |
dce840a0 | 6774 | rcu_read_lock(); |
abcd083a | 6775 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6776 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6777 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6778 | } |
dce840a0 | 6779 | rcu_read_unlock(); |
51888ca2 | 6780 | |
822ff793 | 6781 | ret = 0; |
51888ca2 | 6782 | error: |
2109b99e | 6783 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6784 | return ret; |
1da177e4 | 6785 | } |
029190c5 | 6786 | |
acc3f5d7 | 6787 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6788 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6789 | static struct sched_domain_attr *dattr_cur; |
6790 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6791 | |
6792 | /* | |
6793 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6794 | * cpumask) fails, then fallback to a single sched domain, |
6795 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6796 | */ |
4212823f | 6797 | static cpumask_var_t fallback_doms; |
029190c5 | 6798 | |
ee79d1bd HC |
6799 | /* |
6800 | * arch_update_cpu_topology lets virtualized architectures update the | |
6801 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6802 | * or 0 if it stayed the same. | |
6803 | */ | |
52f5684c | 6804 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6805 | { |
ee79d1bd | 6806 | return 0; |
22e52b07 HC |
6807 | } |
6808 | ||
acc3f5d7 RR |
6809 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6810 | { | |
6811 | int i; | |
6812 | cpumask_var_t *doms; | |
6813 | ||
6814 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6815 | if (!doms) | |
6816 | return NULL; | |
6817 | for (i = 0; i < ndoms; i++) { | |
6818 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6819 | free_sched_domains(doms, i); | |
6820 | return NULL; | |
6821 | } | |
6822 | } | |
6823 | return doms; | |
6824 | } | |
6825 | ||
6826 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6827 | { | |
6828 | unsigned int i; | |
6829 | for (i = 0; i < ndoms; i++) | |
6830 | free_cpumask_var(doms[i]); | |
6831 | kfree(doms); | |
6832 | } | |
6833 | ||
1a20ff27 | 6834 | /* |
41a2d6cf | 6835 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6836 | * For now this just excludes isolated cpus, but could be used to |
6837 | * exclude other special cases in the future. | |
1a20ff27 | 6838 | */ |
c4a8849a | 6839 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6840 | { |
7378547f MM |
6841 | int err; |
6842 | ||
22e52b07 | 6843 | arch_update_cpu_topology(); |
029190c5 | 6844 | ndoms_cur = 1; |
acc3f5d7 | 6845 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6846 | if (!doms_cur) |
acc3f5d7 RR |
6847 | doms_cur = &fallback_doms; |
6848 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6849 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6850 | register_sched_domain_sysctl(); |
7378547f MM |
6851 | |
6852 | return err; | |
1a20ff27 DG |
6853 | } |
6854 | ||
1a20ff27 DG |
6855 | /* |
6856 | * Detach sched domains from a group of cpus specified in cpu_map | |
6857 | * These cpus will now be attached to the NULL domain | |
6858 | */ | |
96f874e2 | 6859 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6860 | { |
6861 | int i; | |
6862 | ||
dce840a0 | 6863 | rcu_read_lock(); |
abcd083a | 6864 | for_each_cpu(i, cpu_map) |
57d885fe | 6865 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6866 | rcu_read_unlock(); |
1a20ff27 DG |
6867 | } |
6868 | ||
1d3504fc HS |
6869 | /* handle null as "default" */ |
6870 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6871 | struct sched_domain_attr *new, int idx_new) | |
6872 | { | |
6873 | struct sched_domain_attr tmp; | |
6874 | ||
6875 | /* fast path */ | |
6876 | if (!new && !cur) | |
6877 | return 1; | |
6878 | ||
6879 | tmp = SD_ATTR_INIT; | |
6880 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6881 | new ? (new + idx_new) : &tmp, | |
6882 | sizeof(struct sched_domain_attr)); | |
6883 | } | |
6884 | ||
029190c5 PJ |
6885 | /* |
6886 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6887 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6888 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6889 | * It destroys each deleted domain and builds each new domain. | |
6890 | * | |
acc3f5d7 | 6891 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6892 | * The masks don't intersect (don't overlap.) We should setup one |
6893 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6894 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6895 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6896 | * it as it is. | |
6897 | * | |
acc3f5d7 RR |
6898 | * The passed in 'doms_new' should be allocated using |
6899 | * alloc_sched_domains. This routine takes ownership of it and will | |
6900 | * free_sched_domains it when done with it. If the caller failed the | |
6901 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6902 | * and partition_sched_domains() will fallback to the single partition | |
6903 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6904 | * |
96f874e2 | 6905 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6906 | * ndoms_new == 0 is a special case for destroying existing domains, |
6907 | * and it will not create the default domain. | |
dfb512ec | 6908 | * |
029190c5 PJ |
6909 | * Call with hotplug lock held |
6910 | */ | |
acc3f5d7 | 6911 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6912 | struct sched_domain_attr *dattr_new) |
029190c5 | 6913 | { |
dfb512ec | 6914 | int i, j, n; |
d65bd5ec | 6915 | int new_topology; |
029190c5 | 6916 | |
712555ee | 6917 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6918 | |
7378547f MM |
6919 | /* always unregister in case we don't destroy any domains */ |
6920 | unregister_sched_domain_sysctl(); | |
6921 | ||
d65bd5ec HC |
6922 | /* Let architecture update cpu core mappings. */ |
6923 | new_topology = arch_update_cpu_topology(); | |
6924 | ||
dfb512ec | 6925 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6926 | |
6927 | /* Destroy deleted domains */ | |
6928 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6929 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6930 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6931 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6932 | goto match1; |
6933 | } | |
6934 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6935 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6936 | match1: |
6937 | ; | |
6938 | } | |
6939 | ||
c8d2d47a | 6940 | n = ndoms_cur; |
e761b772 | 6941 | if (doms_new == NULL) { |
c8d2d47a | 6942 | n = 0; |
acc3f5d7 | 6943 | doms_new = &fallback_doms; |
6ad4c188 | 6944 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6945 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6946 | } |
6947 | ||
029190c5 PJ |
6948 | /* Build new domains */ |
6949 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6950 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6951 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6952 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6953 | goto match2; |
6954 | } | |
6955 | /* no match - add a new doms_new */ | |
dce840a0 | 6956 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6957 | match2: |
6958 | ; | |
6959 | } | |
6960 | ||
6961 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6962 | if (doms_cur != &fallback_doms) |
6963 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6964 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6965 | doms_cur = doms_new; |
1d3504fc | 6966 | dattr_cur = dattr_new; |
029190c5 | 6967 | ndoms_cur = ndoms_new; |
7378547f MM |
6968 | |
6969 | register_sched_domain_sysctl(); | |
a1835615 | 6970 | |
712555ee | 6971 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6972 | } |
6973 | ||
d35be8ba SB |
6974 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6975 | ||
1da177e4 | 6976 | /* |
3a101d05 TH |
6977 | * Update cpusets according to cpu_active mask. If cpusets are |
6978 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6979 | * around partition_sched_domains(). | |
d35be8ba SB |
6980 | * |
6981 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6982 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6983 | */ |
40190a78 | 6984 | static void cpuset_cpu_active(void) |
e761b772 | 6985 | { |
40190a78 | 6986 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
6987 | /* |
6988 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6989 | * resume sequence. As long as this is not the last online | |
6990 | * operation in the resume sequence, just build a single sched | |
6991 | * domain, ignoring cpusets. | |
6992 | */ | |
6993 | num_cpus_frozen--; | |
6994 | if (likely(num_cpus_frozen)) { | |
6995 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 6996 | return; |
d35be8ba | 6997 | } |
d35be8ba SB |
6998 | /* |
6999 | * This is the last CPU online operation. So fall through and | |
7000 | * restore the original sched domains by considering the | |
7001 | * cpuset configurations. | |
7002 | */ | |
3a101d05 | 7003 | } |
135fb3e1 | 7004 | cpuset_update_active_cpus(true); |
3a101d05 | 7005 | } |
e761b772 | 7006 | |
40190a78 | 7007 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7008 | { |
3c18d447 | 7009 | unsigned long flags; |
3c18d447 | 7010 | struct dl_bw *dl_b; |
533445c6 OS |
7011 | bool overflow; |
7012 | int cpus; | |
3c18d447 | 7013 | |
40190a78 | 7014 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
7015 | rcu_read_lock_sched(); |
7016 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7017 | |
533445c6 OS |
7018 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7019 | cpus = dl_bw_cpus(cpu); | |
7020 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7021 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7022 | |
533445c6 | 7023 | rcu_read_unlock_sched(); |
3c18d447 | 7024 | |
533445c6 | 7025 | if (overflow) |
135fb3e1 | 7026 | return -EBUSY; |
7ddf96b0 | 7027 | cpuset_update_active_cpus(false); |
135fb3e1 | 7028 | } else { |
d35be8ba SB |
7029 | num_cpus_frozen++; |
7030 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 TG |
7031 | } |
7032 | return 0; | |
7033 | } | |
7034 | ||
40190a78 | 7035 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7036 | { |
7d976699 TG |
7037 | struct rq *rq = cpu_rq(cpu); |
7038 | unsigned long flags; | |
7039 | ||
40190a78 | 7040 | set_cpu_active(cpu, true); |
135fb3e1 | 7041 | |
40190a78 | 7042 | if (sched_smp_initialized) { |
135fb3e1 | 7043 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7044 | cpuset_cpu_active(); |
e761b772 | 7045 | } |
7d976699 TG |
7046 | |
7047 | /* | |
7048 | * Put the rq online, if not already. This happens: | |
7049 | * | |
7050 | * 1) In the early boot process, because we build the real domains | |
7051 | * after all cpus have been brought up. | |
7052 | * | |
7053 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7054 | * domains. | |
7055 | */ | |
7056 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7057 | if (rq->rd) { | |
7058 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7059 | set_rq_online(rq); | |
7060 | } | |
7061 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7062 | ||
7063 | update_max_interval(); | |
7064 | ||
40190a78 | 7065 | return 0; |
135fb3e1 TG |
7066 | } |
7067 | ||
40190a78 | 7068 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7069 | { |
135fb3e1 TG |
7070 | int ret; |
7071 | ||
40190a78 | 7072 | set_cpu_active(cpu, false); |
b2454caa PZ |
7073 | /* |
7074 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7075 | * users of this state to go away such that all new such users will | |
7076 | * observe it. | |
7077 | * | |
7078 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
7079 | * not imply sync_sched(), so wait for both. | |
7080 | * | |
7081 | * Do sync before park smpboot threads to take care the rcu boost case. | |
7082 | */ | |
7083 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
7084 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
7085 | else | |
7086 | synchronize_rcu(); | |
40190a78 TG |
7087 | |
7088 | if (!sched_smp_initialized) | |
7089 | return 0; | |
7090 | ||
7091 | ret = cpuset_cpu_inactive(cpu); | |
7092 | if (ret) { | |
7093 | set_cpu_active(cpu, true); | |
7094 | return ret; | |
135fb3e1 | 7095 | } |
40190a78 TG |
7096 | sched_domains_numa_masks_clear(cpu); |
7097 | return 0; | |
135fb3e1 TG |
7098 | } |
7099 | ||
94baf7a5 TG |
7100 | static void sched_rq_cpu_starting(unsigned int cpu) |
7101 | { | |
7102 | struct rq *rq = cpu_rq(cpu); | |
7103 | ||
7104 | rq->calc_load_update = calc_load_update; | |
7105 | account_reset_rq(rq); | |
7106 | update_max_interval(); | |
7107 | } | |
7108 | ||
135fb3e1 TG |
7109 | int sched_cpu_starting(unsigned int cpu) |
7110 | { | |
7111 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 7112 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 7113 | return 0; |
e761b772 | 7114 | } |
e761b772 | 7115 | |
f2785ddb TG |
7116 | #ifdef CONFIG_HOTPLUG_CPU |
7117 | int sched_cpu_dying(unsigned int cpu) | |
7118 | { | |
7119 | struct rq *rq = cpu_rq(cpu); | |
7120 | unsigned long flags; | |
7121 | ||
7122 | /* Handle pending wakeups and then migrate everything off */ | |
7123 | sched_ttwu_pending(); | |
7124 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7125 | if (rq->rd) { | |
7126 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7127 | set_rq_offline(rq); | |
7128 | } | |
7129 | migrate_tasks(rq); | |
7130 | BUG_ON(rq->nr_running != 1); | |
7131 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7132 | calc_load_migrate(rq); | |
7133 | update_max_interval(); | |
20a5c8cc | 7134 | nohz_balance_exit_idle(cpu); |
f2785ddb TG |
7135 | return 0; |
7136 | } | |
7137 | #endif | |
7138 | ||
1da177e4 LT |
7139 | void __init sched_init_smp(void) |
7140 | { | |
dcc30a35 RR |
7141 | cpumask_var_t non_isolated_cpus; |
7142 | ||
7143 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7144 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7145 | |
cb83b629 PZ |
7146 | sched_init_numa(); |
7147 | ||
6acce3ef PZ |
7148 | /* |
7149 | * There's no userspace yet to cause hotplug operations; hence all the | |
7150 | * cpu masks are stable and all blatant races in the below code cannot | |
7151 | * happen. | |
7152 | */ | |
712555ee | 7153 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7154 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7155 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7156 | if (cpumask_empty(non_isolated_cpus)) | |
7157 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7158 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7159 | |
b328ca18 | 7160 | init_hrtick(); |
5c1e1767 NP |
7161 | |
7162 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7163 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7164 | BUG(); |
19978ca6 | 7165 | sched_init_granularity(); |
dcc30a35 | 7166 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7167 | |
0e3900e6 | 7168 | init_sched_rt_class(); |
1baca4ce | 7169 | init_sched_dl_class(); |
e26fbffd | 7170 | sched_smp_initialized = true; |
1da177e4 | 7171 | } |
e26fbffd TG |
7172 | |
7173 | static int __init migration_init(void) | |
7174 | { | |
94baf7a5 | 7175 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd TG |
7176 | return 0; |
7177 | } | |
7178 | early_initcall(migration_init); | |
7179 | ||
1da177e4 LT |
7180 | #else |
7181 | void __init sched_init_smp(void) | |
7182 | { | |
19978ca6 | 7183 | sched_init_granularity(); |
1da177e4 LT |
7184 | } |
7185 | #endif /* CONFIG_SMP */ | |
7186 | ||
7187 | int in_sched_functions(unsigned long addr) | |
7188 | { | |
1da177e4 LT |
7189 | return in_lock_functions(addr) || |
7190 | (addr >= (unsigned long)__sched_text_start | |
7191 | && addr < (unsigned long)__sched_text_end); | |
7192 | } | |
7193 | ||
029632fb | 7194 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7195 | /* |
7196 | * Default task group. | |
7197 | * Every task in system belongs to this group at bootup. | |
7198 | */ | |
029632fb | 7199 | struct task_group root_task_group; |
35cf4e50 | 7200 | LIST_HEAD(task_groups); |
b0367629 WL |
7201 | |
7202 | /* Cacheline aligned slab cache for task_group */ | |
7203 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7204 | #endif |
6f505b16 | 7205 | |
e6252c3e | 7206 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7207 | |
1da177e4 LT |
7208 | void __init sched_init(void) |
7209 | { | |
dd41f596 | 7210 | int i, j; |
434d53b0 MT |
7211 | unsigned long alloc_size = 0, ptr; |
7212 | ||
7213 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7214 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7215 | #endif | |
7216 | #ifdef CONFIG_RT_GROUP_SCHED | |
7217 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7218 | #endif | |
434d53b0 | 7219 | if (alloc_size) { |
36b7b6d4 | 7220 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7221 | |
7222 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7223 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7224 | ptr += nr_cpu_ids * sizeof(void **); |
7225 | ||
07e06b01 | 7226 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7227 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7228 | |
6d6bc0ad | 7229 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7230 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7231 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7232 | ptr += nr_cpu_ids * sizeof(void **); |
7233 | ||
07e06b01 | 7234 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7235 | ptr += nr_cpu_ids * sizeof(void **); |
7236 | ||
6d6bc0ad | 7237 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7238 | } |
df7c8e84 | 7239 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7240 | for_each_possible_cpu(i) { |
7241 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7242 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7243 | } |
b74e6278 | 7244 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7245 | |
332ac17e DF |
7246 | init_rt_bandwidth(&def_rt_bandwidth, |
7247 | global_rt_period(), global_rt_runtime()); | |
7248 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7249 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7250 | |
57d885fe GH |
7251 | #ifdef CONFIG_SMP |
7252 | init_defrootdomain(); | |
7253 | #endif | |
7254 | ||
d0b27fa7 | 7255 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7256 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7257 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7258 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7259 | |
7c941438 | 7260 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7261 | task_group_cache = KMEM_CACHE(task_group, 0); |
7262 | ||
07e06b01 YZ |
7263 | list_add(&root_task_group.list, &task_groups); |
7264 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7265 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7266 | autogroup_init(&init_task); |
7c941438 | 7267 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7268 | |
0a945022 | 7269 | for_each_possible_cpu(i) { |
70b97a7f | 7270 | struct rq *rq; |
1da177e4 LT |
7271 | |
7272 | rq = cpu_rq(i); | |
05fa785c | 7273 | raw_spin_lock_init(&rq->lock); |
7897986b | 7274 | rq->nr_running = 0; |
dce48a84 TG |
7275 | rq->calc_load_active = 0; |
7276 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7277 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7278 | init_rt_rq(&rq->rt); |
7279 | init_dl_rq(&rq->dl); | |
dd41f596 | 7280 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7281 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7282 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7283 | /* |
07e06b01 | 7284 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7285 | * |
7286 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7287 | * gets 100% of the cpu resources in the system. This overall | |
7288 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7289 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7290 | * based on each entity's (task or task-group's) weight |
7291 | * (se->load.weight). | |
7292 | * | |
07e06b01 | 7293 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7294 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7295 | * then A0's share of the cpu resource is: | |
7296 | * | |
0d905bca | 7297 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7298 | * |
07e06b01 YZ |
7299 | * We achieve this by letting root_task_group's tasks sit |
7300 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7301 | */ |
ab84d31e | 7302 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7303 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7304 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7305 | ||
7306 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7307 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7308 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7309 | #endif |
1da177e4 | 7310 | |
dd41f596 IM |
7311 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7312 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7313 | |
7314 | rq->last_load_update_tick = jiffies; | |
7315 | ||
1da177e4 | 7316 | #ifdef CONFIG_SMP |
41c7ce9a | 7317 | rq->sd = NULL; |
57d885fe | 7318 | rq->rd = NULL; |
ca6d75e6 | 7319 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7320 | rq->balance_callback = NULL; |
1da177e4 | 7321 | rq->active_balance = 0; |
dd41f596 | 7322 | rq->next_balance = jiffies; |
1da177e4 | 7323 | rq->push_cpu = 0; |
0a2966b4 | 7324 | rq->cpu = i; |
1f11eb6a | 7325 | rq->online = 0; |
eae0c9df MG |
7326 | rq->idle_stamp = 0; |
7327 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7328 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7329 | |
7330 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7331 | ||
dc938520 | 7332 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7333 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7334 | rq->nohz_flags = 0; |
83cd4fe2 | 7335 | #endif |
265f22a9 FW |
7336 | #ifdef CONFIG_NO_HZ_FULL |
7337 | rq->last_sched_tick = 0; | |
7338 | #endif | |
1da177e4 | 7339 | #endif |
8f4d37ec | 7340 | init_rq_hrtick(rq); |
1da177e4 | 7341 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7342 | } |
7343 | ||
2dd73a4f | 7344 | set_load_weight(&init_task); |
b50f60ce | 7345 | |
e107be36 AK |
7346 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7347 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7348 | #endif | |
7349 | ||
1da177e4 LT |
7350 | /* |
7351 | * The boot idle thread does lazy MMU switching as well: | |
7352 | */ | |
7353 | atomic_inc(&init_mm.mm_count); | |
7354 | enter_lazy_tlb(&init_mm, current); | |
7355 | ||
1b537c7d YD |
7356 | /* |
7357 | * During early bootup we pretend to be a normal task: | |
7358 | */ | |
7359 | current->sched_class = &fair_sched_class; | |
7360 | ||
1da177e4 LT |
7361 | /* |
7362 | * Make us the idle thread. Technically, schedule() should not be | |
7363 | * called from this thread, however somewhere below it might be, | |
7364 | * but because we are the idle thread, we just pick up running again | |
7365 | * when this runqueue becomes "idle". | |
7366 | */ | |
7367 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7368 | |
7369 | calc_load_update = jiffies + LOAD_FREQ; | |
7370 | ||
bf4d83f6 | 7371 | #ifdef CONFIG_SMP |
4cb98839 | 7372 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7373 | /* May be allocated at isolcpus cmdline parse time */ |
7374 | if (cpu_isolated_map == NULL) | |
7375 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7376 | idle_thread_set_boot_cpu(); |
9cf7243d | 7377 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
7378 | #endif |
7379 | init_sched_fair_class(); | |
6a7b3dc3 | 7380 | |
6892b75e | 7381 | scheduler_running = 1; |
1da177e4 LT |
7382 | } |
7383 | ||
d902db1e | 7384 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7385 | static inline int preempt_count_equals(int preempt_offset) |
7386 | { | |
da7142e2 | 7387 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7388 | |
4ba8216c | 7389 | return (nested == preempt_offset); |
e4aafea2 FW |
7390 | } |
7391 | ||
d894837f | 7392 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7393 | { |
8eb23b9f PZ |
7394 | /* |
7395 | * Blocking primitives will set (and therefore destroy) current->state, | |
7396 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7397 | * otherwise we will destroy state. | |
7398 | */ | |
00845eb9 | 7399 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7400 | "do not call blocking ops when !TASK_RUNNING; " |
7401 | "state=%lx set at [<%p>] %pS\n", | |
7402 | current->state, | |
7403 | (void *)current->task_state_change, | |
00845eb9 | 7404 | (void *)current->task_state_change); |
8eb23b9f | 7405 | |
3427445a PZ |
7406 | ___might_sleep(file, line, preempt_offset); |
7407 | } | |
7408 | EXPORT_SYMBOL(__might_sleep); | |
7409 | ||
7410 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7411 | { |
1da177e4 LT |
7412 | static unsigned long prev_jiffy; /* ratelimiting */ |
7413 | ||
b3fbab05 | 7414 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7415 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7416 | !is_idle_task(current)) || | |
e4aafea2 | 7417 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7418 | return; |
7419 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7420 | return; | |
7421 | prev_jiffy = jiffies; | |
7422 | ||
3df0fc5b PZ |
7423 | printk(KERN_ERR |
7424 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7425 | file, line); | |
7426 | printk(KERN_ERR | |
7427 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7428 | in_atomic(), irqs_disabled(), | |
7429 | current->pid, current->comm); | |
aef745fc | 7430 | |
a8b686b3 ES |
7431 | if (task_stack_end_corrupted(current)) |
7432 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7433 | ||
aef745fc IM |
7434 | debug_show_held_locks(current); |
7435 | if (irqs_disabled()) | |
7436 | print_irqtrace_events(current); | |
8f47b187 TG |
7437 | #ifdef CONFIG_DEBUG_PREEMPT |
7438 | if (!preempt_count_equals(preempt_offset)) { | |
7439 | pr_err("Preemption disabled at:"); | |
7440 | print_ip_sym(current->preempt_disable_ip); | |
7441 | pr_cont("\n"); | |
7442 | } | |
7443 | #endif | |
aef745fc | 7444 | dump_stack(); |
1da177e4 | 7445 | } |
3427445a | 7446 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7447 | #endif |
7448 | ||
7449 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7450 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7451 | { |
dbc7f069 | 7452 | struct task_struct *g, *p; |
d50dde5a DF |
7453 | struct sched_attr attr = { |
7454 | .sched_policy = SCHED_NORMAL, | |
7455 | }; | |
1da177e4 | 7456 | |
3472eaa1 | 7457 | read_lock(&tasklist_lock); |
5d07f420 | 7458 | for_each_process_thread(g, p) { |
178be793 IM |
7459 | /* |
7460 | * Only normalize user tasks: | |
7461 | */ | |
3472eaa1 | 7462 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7463 | continue; |
7464 | ||
6cfb0d5d | 7465 | p->se.exec_start = 0; |
6cfb0d5d | 7466 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7467 | p->se.statistics.wait_start = 0; |
7468 | p->se.statistics.sleep_start = 0; | |
7469 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7470 | #endif |
dd41f596 | 7471 | |
aab03e05 | 7472 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7473 | /* |
7474 | * Renice negative nice level userspace | |
7475 | * tasks back to 0: | |
7476 | */ | |
3472eaa1 | 7477 | if (task_nice(p) < 0) |
dd41f596 | 7478 | set_user_nice(p, 0); |
1da177e4 | 7479 | continue; |
dd41f596 | 7480 | } |
1da177e4 | 7481 | |
dbc7f069 | 7482 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7483 | } |
3472eaa1 | 7484 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7485 | } |
7486 | ||
7487 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7488 | |
67fc4e0c | 7489 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7490 | /* |
67fc4e0c | 7491 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7492 | * |
7493 | * They can only be called when the whole system has been | |
7494 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7495 | * activity can take place. Using them for anything else would | |
7496 | * be a serious bug, and as a result, they aren't even visible | |
7497 | * under any other configuration. | |
7498 | */ | |
7499 | ||
7500 | /** | |
7501 | * curr_task - return the current task for a given cpu. | |
7502 | * @cpu: the processor in question. | |
7503 | * | |
7504 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7505 | * |
7506 | * Return: The current task for @cpu. | |
1df5c10a | 7507 | */ |
36c8b586 | 7508 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7509 | { |
7510 | return cpu_curr(cpu); | |
7511 | } | |
7512 | ||
67fc4e0c JW |
7513 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7514 | ||
7515 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7516 | /** |
7517 | * set_curr_task - set the current task for a given cpu. | |
7518 | * @cpu: the processor in question. | |
7519 | * @p: the task pointer to set. | |
7520 | * | |
7521 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7522 | * are serviced on a separate stack. It allows the architecture to switch the |
7523 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7524 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7525 | * and caller must save the original value of the current task (see | |
7526 | * curr_task() above) and restore that value before reenabling interrupts and | |
7527 | * re-starting the system. | |
7528 | * | |
7529 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7530 | */ | |
36c8b586 | 7531 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7532 | { |
7533 | cpu_curr(cpu) = p; | |
7534 | } | |
7535 | ||
7536 | #endif | |
29f59db3 | 7537 | |
7c941438 | 7538 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7539 | /* task_group_lock serializes the addition/removal of task groups */ |
7540 | static DEFINE_SPINLOCK(task_group_lock); | |
7541 | ||
2f5177f0 | 7542 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7543 | { |
7544 | free_fair_sched_group(tg); | |
7545 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7546 | autogroup_free(tg); |
b0367629 | 7547 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7548 | } |
7549 | ||
7550 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7551 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7552 | { |
7553 | struct task_group *tg; | |
bccbe08a | 7554 | |
b0367629 | 7555 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7556 | if (!tg) |
7557 | return ERR_PTR(-ENOMEM); | |
7558 | ||
ec7dc8ac | 7559 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7560 | goto err; |
7561 | ||
ec7dc8ac | 7562 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7563 | goto err; |
7564 | ||
ace783b9 LZ |
7565 | return tg; |
7566 | ||
7567 | err: | |
2f5177f0 | 7568 | sched_free_group(tg); |
ace783b9 LZ |
7569 | return ERR_PTR(-ENOMEM); |
7570 | } | |
7571 | ||
7572 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7573 | { | |
7574 | unsigned long flags; | |
7575 | ||
8ed36996 | 7576 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7577 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7578 | |
7579 | WARN_ON(!parent); /* root should already exist */ | |
7580 | ||
7581 | tg->parent = parent; | |
f473aa5e | 7582 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7583 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7584 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7585 | } |
7586 | ||
9b5b7751 | 7587 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7588 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7589 | { |
29f59db3 | 7590 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 7591 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7592 | } |
7593 | ||
4cf86d77 | 7594 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7595 | { |
7596 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 7597 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7598 | } |
7599 | ||
7600 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7601 | { |
8ed36996 | 7602 | unsigned long flags; |
29f59db3 | 7603 | |
3d4b47b4 | 7604 | /* end participation in shares distribution */ |
6fe1f348 | 7605 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7606 | |
7607 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7608 | list_del_rcu(&tg->list); |
f473aa5e | 7609 | list_del_rcu(&tg->siblings); |
8ed36996 | 7610 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7611 | } |
7612 | ||
9b5b7751 | 7613 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7614 | * The caller of this function should have put the task in its new group |
7615 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7616 | * reflect its new group. | |
9b5b7751 SV |
7617 | */ |
7618 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7619 | { |
8323f26c | 7620 | struct task_group *tg; |
da0c1e65 | 7621 | int queued, running; |
29f59db3 SV |
7622 | unsigned long flags; |
7623 | struct rq *rq; | |
7624 | ||
7625 | rq = task_rq_lock(tsk, &flags); | |
7626 | ||
051a1d1a | 7627 | running = task_current(rq, tsk); |
da0c1e65 | 7628 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7629 | |
da0c1e65 | 7630 | if (queued) |
ff77e468 | 7631 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); |
0e1f3483 | 7632 | if (unlikely(running)) |
f3cd1c4e | 7633 | put_prev_task(rq, tsk); |
29f59db3 | 7634 | |
f7b8a47d KT |
7635 | /* |
7636 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7637 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7638 | * to prevent lockdep warnings. | |
7639 | */ | |
7640 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7641 | struct task_group, css); |
7642 | tg = autogroup_task_group(tsk, tg); | |
7643 | tsk->sched_task_group = tg; | |
7644 | ||
810b3817 | 7645 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7646 | if (tsk->sched_class->task_move_group) |
bc54da21 | 7647 | tsk->sched_class->task_move_group(tsk); |
b2b5ce02 | 7648 | else |
810b3817 | 7649 | #endif |
b2b5ce02 | 7650 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7651 | |
0e1f3483 HS |
7652 | if (unlikely(running)) |
7653 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7654 | if (queued) |
ff77e468 | 7655 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
29f59db3 | 7656 | |
0122ec5b | 7657 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7658 | } |
7c941438 | 7659 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7660 | |
a790de99 PT |
7661 | #ifdef CONFIG_RT_GROUP_SCHED |
7662 | /* | |
7663 | * Ensure that the real time constraints are schedulable. | |
7664 | */ | |
7665 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7666 | |
9a7e0b18 PZ |
7667 | /* Must be called with tasklist_lock held */ |
7668 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7669 | { |
9a7e0b18 | 7670 | struct task_struct *g, *p; |
b40b2e8e | 7671 | |
1fe89e1b PZ |
7672 | /* |
7673 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7674 | */ | |
7675 | if (task_group_is_autogroup(tg)) | |
7676 | return 0; | |
7677 | ||
5d07f420 | 7678 | for_each_process_thread(g, p) { |
8651c658 | 7679 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7680 | return 1; |
5d07f420 | 7681 | } |
b40b2e8e | 7682 | |
9a7e0b18 PZ |
7683 | return 0; |
7684 | } | |
b40b2e8e | 7685 | |
9a7e0b18 PZ |
7686 | struct rt_schedulable_data { |
7687 | struct task_group *tg; | |
7688 | u64 rt_period; | |
7689 | u64 rt_runtime; | |
7690 | }; | |
b40b2e8e | 7691 | |
a790de99 | 7692 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7693 | { |
7694 | struct rt_schedulable_data *d = data; | |
7695 | struct task_group *child; | |
7696 | unsigned long total, sum = 0; | |
7697 | u64 period, runtime; | |
b40b2e8e | 7698 | |
9a7e0b18 PZ |
7699 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7700 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7701 | |
9a7e0b18 PZ |
7702 | if (tg == d->tg) { |
7703 | period = d->rt_period; | |
7704 | runtime = d->rt_runtime; | |
b40b2e8e | 7705 | } |
b40b2e8e | 7706 | |
4653f803 PZ |
7707 | /* |
7708 | * Cannot have more runtime than the period. | |
7709 | */ | |
7710 | if (runtime > period && runtime != RUNTIME_INF) | |
7711 | return -EINVAL; | |
6f505b16 | 7712 | |
4653f803 PZ |
7713 | /* |
7714 | * Ensure we don't starve existing RT tasks. | |
7715 | */ | |
9a7e0b18 PZ |
7716 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7717 | return -EBUSY; | |
6f505b16 | 7718 | |
9a7e0b18 | 7719 | total = to_ratio(period, runtime); |
6f505b16 | 7720 | |
4653f803 PZ |
7721 | /* |
7722 | * Nobody can have more than the global setting allows. | |
7723 | */ | |
7724 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7725 | return -EINVAL; | |
6f505b16 | 7726 | |
4653f803 PZ |
7727 | /* |
7728 | * The sum of our children's runtime should not exceed our own. | |
7729 | */ | |
9a7e0b18 PZ |
7730 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7731 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7732 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7733 | |
9a7e0b18 PZ |
7734 | if (child == d->tg) { |
7735 | period = d->rt_period; | |
7736 | runtime = d->rt_runtime; | |
7737 | } | |
6f505b16 | 7738 | |
9a7e0b18 | 7739 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7740 | } |
6f505b16 | 7741 | |
9a7e0b18 PZ |
7742 | if (sum > total) |
7743 | return -EINVAL; | |
7744 | ||
7745 | return 0; | |
6f505b16 PZ |
7746 | } |
7747 | ||
9a7e0b18 | 7748 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7749 | { |
8277434e PT |
7750 | int ret; |
7751 | ||
9a7e0b18 PZ |
7752 | struct rt_schedulable_data data = { |
7753 | .tg = tg, | |
7754 | .rt_period = period, | |
7755 | .rt_runtime = runtime, | |
7756 | }; | |
7757 | ||
8277434e PT |
7758 | rcu_read_lock(); |
7759 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7760 | rcu_read_unlock(); | |
7761 | ||
7762 | return ret; | |
521f1a24 DG |
7763 | } |
7764 | ||
ab84d31e | 7765 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7766 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7767 | { |
ac086bc2 | 7768 | int i, err = 0; |
9f0c1e56 | 7769 | |
2636ed5f PZ |
7770 | /* |
7771 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7772 | * kernel creating (and or operating) RT threads. | |
7773 | */ | |
7774 | if (tg == &root_task_group && rt_runtime == 0) | |
7775 | return -EINVAL; | |
7776 | ||
7777 | /* No period doesn't make any sense. */ | |
7778 | if (rt_period == 0) | |
7779 | return -EINVAL; | |
7780 | ||
9f0c1e56 | 7781 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7782 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7783 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7784 | if (err) | |
9f0c1e56 | 7785 | goto unlock; |
ac086bc2 | 7786 | |
0986b11b | 7787 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7788 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7789 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7790 | |
7791 | for_each_possible_cpu(i) { | |
7792 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7793 | ||
0986b11b | 7794 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7795 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7796 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7797 | } |
0986b11b | 7798 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7799 | unlock: |
521f1a24 | 7800 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7801 | mutex_unlock(&rt_constraints_mutex); |
7802 | ||
7803 | return err; | |
6f505b16 PZ |
7804 | } |
7805 | ||
25cc7da7 | 7806 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7807 | { |
7808 | u64 rt_runtime, rt_period; | |
7809 | ||
7810 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7811 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7812 | if (rt_runtime_us < 0) | |
7813 | rt_runtime = RUNTIME_INF; | |
7814 | ||
ab84d31e | 7815 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7816 | } |
7817 | ||
25cc7da7 | 7818 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7819 | { |
7820 | u64 rt_runtime_us; | |
7821 | ||
d0b27fa7 | 7822 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7823 | return -1; |
7824 | ||
d0b27fa7 | 7825 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7826 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7827 | return rt_runtime_us; | |
7828 | } | |
d0b27fa7 | 7829 | |
ce2f5fe4 | 7830 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7831 | { |
7832 | u64 rt_runtime, rt_period; | |
7833 | ||
ce2f5fe4 | 7834 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7835 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7836 | ||
ab84d31e | 7837 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7838 | } |
7839 | ||
25cc7da7 | 7840 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7841 | { |
7842 | u64 rt_period_us; | |
7843 | ||
7844 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7845 | do_div(rt_period_us, NSEC_PER_USEC); | |
7846 | return rt_period_us; | |
7847 | } | |
332ac17e | 7848 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7849 | |
332ac17e | 7850 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7851 | static int sched_rt_global_constraints(void) |
7852 | { | |
7853 | int ret = 0; | |
7854 | ||
7855 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7856 | read_lock(&tasklist_lock); |
4653f803 | 7857 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7858 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7859 | mutex_unlock(&rt_constraints_mutex); |
7860 | ||
7861 | return ret; | |
7862 | } | |
54e99124 | 7863 | |
25cc7da7 | 7864 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7865 | { |
7866 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7867 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7868 | return 0; | |
7869 | ||
7870 | return 1; | |
7871 | } | |
7872 | ||
6d6bc0ad | 7873 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7874 | static int sched_rt_global_constraints(void) |
7875 | { | |
ac086bc2 | 7876 | unsigned long flags; |
332ac17e | 7877 | int i, ret = 0; |
ec5d4989 | 7878 | |
0986b11b | 7879 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7880 | for_each_possible_cpu(i) { |
7881 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7882 | ||
0986b11b | 7883 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7884 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7885 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7886 | } |
0986b11b | 7887 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7888 | |
332ac17e | 7889 | return ret; |
d0b27fa7 | 7890 | } |
6d6bc0ad | 7891 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7892 | |
a1963b81 | 7893 | static int sched_dl_global_validate(void) |
332ac17e | 7894 | { |
1724813d PZ |
7895 | u64 runtime = global_rt_runtime(); |
7896 | u64 period = global_rt_period(); | |
332ac17e | 7897 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7898 | struct dl_bw *dl_b; |
1724813d | 7899 | int cpu, ret = 0; |
49516342 | 7900 | unsigned long flags; |
332ac17e DF |
7901 | |
7902 | /* | |
7903 | * Here we want to check the bandwidth not being set to some | |
7904 | * value smaller than the currently allocated bandwidth in | |
7905 | * any of the root_domains. | |
7906 | * | |
7907 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7908 | * cycling on root_domains... Discussion on different/better | |
7909 | * solutions is welcome! | |
7910 | */ | |
1724813d | 7911 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7912 | rcu_read_lock_sched(); |
7913 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7914 | |
49516342 | 7915 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7916 | if (new_bw < dl_b->total_bw) |
7917 | ret = -EBUSY; | |
49516342 | 7918 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7919 | |
f10e00f4 KT |
7920 | rcu_read_unlock_sched(); |
7921 | ||
1724813d PZ |
7922 | if (ret) |
7923 | break; | |
332ac17e DF |
7924 | } |
7925 | ||
1724813d | 7926 | return ret; |
332ac17e DF |
7927 | } |
7928 | ||
1724813d | 7929 | static void sched_dl_do_global(void) |
ce0dbbbb | 7930 | { |
1724813d | 7931 | u64 new_bw = -1; |
f10e00f4 | 7932 | struct dl_bw *dl_b; |
1724813d | 7933 | int cpu; |
49516342 | 7934 | unsigned long flags; |
ce0dbbbb | 7935 | |
1724813d PZ |
7936 | def_dl_bandwidth.dl_period = global_rt_period(); |
7937 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7938 | ||
7939 | if (global_rt_runtime() != RUNTIME_INF) | |
7940 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7941 | ||
7942 | /* | |
7943 | * FIXME: As above... | |
7944 | */ | |
7945 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7946 | rcu_read_lock_sched(); |
7947 | dl_b = dl_bw_of(cpu); | |
1724813d | 7948 | |
49516342 | 7949 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7950 | dl_b->bw = new_bw; |
49516342 | 7951 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
7952 | |
7953 | rcu_read_unlock_sched(); | |
ce0dbbbb | 7954 | } |
1724813d PZ |
7955 | } |
7956 | ||
7957 | static int sched_rt_global_validate(void) | |
7958 | { | |
7959 | if (sysctl_sched_rt_period <= 0) | |
7960 | return -EINVAL; | |
7961 | ||
e9e7cb38 JL |
7962 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7963 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7964 | return -EINVAL; |
7965 | ||
7966 | return 0; | |
7967 | } | |
7968 | ||
7969 | static void sched_rt_do_global(void) | |
7970 | { | |
7971 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7972 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7973 | } |
7974 | ||
d0b27fa7 | 7975 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7976 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7977 | loff_t *ppos) |
7978 | { | |
d0b27fa7 PZ |
7979 | int old_period, old_runtime; |
7980 | static DEFINE_MUTEX(mutex); | |
1724813d | 7981 | int ret; |
d0b27fa7 PZ |
7982 | |
7983 | mutex_lock(&mutex); | |
7984 | old_period = sysctl_sched_rt_period; | |
7985 | old_runtime = sysctl_sched_rt_runtime; | |
7986 | ||
8d65af78 | 7987 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7988 | |
7989 | if (!ret && write) { | |
1724813d PZ |
7990 | ret = sched_rt_global_validate(); |
7991 | if (ret) | |
7992 | goto undo; | |
7993 | ||
a1963b81 | 7994 | ret = sched_dl_global_validate(); |
1724813d PZ |
7995 | if (ret) |
7996 | goto undo; | |
7997 | ||
a1963b81 | 7998 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7999 | if (ret) |
8000 | goto undo; | |
8001 | ||
8002 | sched_rt_do_global(); | |
8003 | sched_dl_do_global(); | |
8004 | } | |
8005 | if (0) { | |
8006 | undo: | |
8007 | sysctl_sched_rt_period = old_period; | |
8008 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8009 | } |
8010 | mutex_unlock(&mutex); | |
8011 | ||
8012 | return ret; | |
8013 | } | |
68318b8e | 8014 | |
1724813d | 8015 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8016 | void __user *buffer, size_t *lenp, |
8017 | loff_t *ppos) | |
8018 | { | |
8019 | int ret; | |
332ac17e | 8020 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8021 | |
8022 | mutex_lock(&mutex); | |
332ac17e | 8023 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8024 | /* make sure that internally we keep jiffies */ |
8025 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8026 | if (!ret && write) { |
1724813d PZ |
8027 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8028 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8029 | } |
8030 | mutex_unlock(&mutex); | |
332ac17e DF |
8031 | return ret; |
8032 | } | |
8033 | ||
052f1dc7 | 8034 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8035 | |
a7c6d554 | 8036 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8037 | { |
a7c6d554 | 8038 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8039 | } |
8040 | ||
eb95419b TH |
8041 | static struct cgroup_subsys_state * |
8042 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8043 | { |
eb95419b TH |
8044 | struct task_group *parent = css_tg(parent_css); |
8045 | struct task_group *tg; | |
68318b8e | 8046 | |
eb95419b | 8047 | if (!parent) { |
68318b8e | 8048 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8049 | return &root_task_group.css; |
68318b8e SV |
8050 | } |
8051 | ||
ec7dc8ac | 8052 | tg = sched_create_group(parent); |
68318b8e SV |
8053 | if (IS_ERR(tg)) |
8054 | return ERR_PTR(-ENOMEM); | |
8055 | ||
2f5177f0 PZ |
8056 | sched_online_group(tg, parent); |
8057 | ||
68318b8e SV |
8058 | return &tg->css; |
8059 | } | |
8060 | ||
2f5177f0 | 8061 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8062 | { |
eb95419b | 8063 | struct task_group *tg = css_tg(css); |
ace783b9 | 8064 | |
2f5177f0 | 8065 | sched_offline_group(tg); |
ace783b9 LZ |
8066 | } |
8067 | ||
eb95419b | 8068 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8069 | { |
eb95419b | 8070 | struct task_group *tg = css_tg(css); |
68318b8e | 8071 | |
2f5177f0 PZ |
8072 | /* |
8073 | * Relies on the RCU grace period between css_released() and this. | |
8074 | */ | |
8075 | sched_free_group(tg); | |
ace783b9 LZ |
8076 | } |
8077 | ||
b53202e6 | 8078 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 KT |
8079 | { |
8080 | sched_move_task(task); | |
8081 | } | |
8082 | ||
1f7dd3e5 | 8083 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8084 | { |
bb9d97b6 | 8085 | struct task_struct *task; |
1f7dd3e5 | 8086 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8087 | |
1f7dd3e5 | 8088 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8089 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8090 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8091 | return -EINVAL; |
b68aa230 | 8092 | #else |
bb9d97b6 TH |
8093 | /* We don't support RT-tasks being in separate groups */ |
8094 | if (task->sched_class != &fair_sched_class) | |
8095 | return -EINVAL; | |
b68aa230 | 8096 | #endif |
bb9d97b6 | 8097 | } |
be367d09 BB |
8098 | return 0; |
8099 | } | |
68318b8e | 8100 | |
1f7dd3e5 | 8101 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8102 | { |
bb9d97b6 | 8103 | struct task_struct *task; |
1f7dd3e5 | 8104 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8105 | |
1f7dd3e5 | 8106 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8107 | sched_move_task(task); |
68318b8e SV |
8108 | } |
8109 | ||
052f1dc7 | 8110 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8111 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8112 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8113 | { |
182446d0 | 8114 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8115 | } |
8116 | ||
182446d0 TH |
8117 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8118 | struct cftype *cft) | |
68318b8e | 8119 | { |
182446d0 | 8120 | struct task_group *tg = css_tg(css); |
68318b8e | 8121 | |
c8b28116 | 8122 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8123 | } |
ab84d31e PT |
8124 | |
8125 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8126 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8127 | ||
ab84d31e PT |
8128 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8129 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8130 | ||
a790de99 PT |
8131 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8132 | ||
ab84d31e PT |
8133 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8134 | { | |
56f570e5 | 8135 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8136 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8137 | |
8138 | if (tg == &root_task_group) | |
8139 | return -EINVAL; | |
8140 | ||
8141 | /* | |
8142 | * Ensure we have at some amount of bandwidth every period. This is | |
8143 | * to prevent reaching a state of large arrears when throttled via | |
8144 | * entity_tick() resulting in prolonged exit starvation. | |
8145 | */ | |
8146 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8147 | return -EINVAL; | |
8148 | ||
8149 | /* | |
8150 | * Likewise, bound things on the otherside by preventing insane quota | |
8151 | * periods. This also allows us to normalize in computing quota | |
8152 | * feasibility. | |
8153 | */ | |
8154 | if (period > max_cfs_quota_period) | |
8155 | return -EINVAL; | |
8156 | ||
0e59bdae KT |
8157 | /* |
8158 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8159 | * unthrottle_offline_cfs_rqs(). | |
8160 | */ | |
8161 | get_online_cpus(); | |
a790de99 PT |
8162 | mutex_lock(&cfs_constraints_mutex); |
8163 | ret = __cfs_schedulable(tg, period, quota); | |
8164 | if (ret) | |
8165 | goto out_unlock; | |
8166 | ||
58088ad0 | 8167 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8168 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8169 | /* |
8170 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8171 | * before making related changes, and on->off must occur afterwards | |
8172 | */ | |
8173 | if (runtime_enabled && !runtime_was_enabled) | |
8174 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8175 | raw_spin_lock_irq(&cfs_b->lock); |
8176 | cfs_b->period = ns_to_ktime(period); | |
8177 | cfs_b->quota = quota; | |
58088ad0 | 8178 | |
a9cf55b2 | 8179 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8180 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8181 | if (runtime_enabled) |
8182 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8183 | raw_spin_unlock_irq(&cfs_b->lock); |
8184 | ||
0e59bdae | 8185 | for_each_online_cpu(i) { |
ab84d31e | 8186 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8187 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8188 | |
8189 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8190 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8191 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8192 | |
029632fb | 8193 | if (cfs_rq->throttled) |
671fd9da | 8194 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8195 | raw_spin_unlock_irq(&rq->lock); |
8196 | } | |
1ee14e6c BS |
8197 | if (runtime_was_enabled && !runtime_enabled) |
8198 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8199 | out_unlock: |
8200 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8201 | put_online_cpus(); |
ab84d31e | 8202 | |
a790de99 | 8203 | return ret; |
ab84d31e PT |
8204 | } |
8205 | ||
8206 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8207 | { | |
8208 | u64 quota, period; | |
8209 | ||
029632fb | 8210 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8211 | if (cfs_quota_us < 0) |
8212 | quota = RUNTIME_INF; | |
8213 | else | |
8214 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8215 | ||
8216 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8217 | } | |
8218 | ||
8219 | long tg_get_cfs_quota(struct task_group *tg) | |
8220 | { | |
8221 | u64 quota_us; | |
8222 | ||
029632fb | 8223 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8224 | return -1; |
8225 | ||
029632fb | 8226 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8227 | do_div(quota_us, NSEC_PER_USEC); |
8228 | ||
8229 | return quota_us; | |
8230 | } | |
8231 | ||
8232 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8233 | { | |
8234 | u64 quota, period; | |
8235 | ||
8236 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8237 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8238 | |
ab84d31e PT |
8239 | return tg_set_cfs_bandwidth(tg, period, quota); |
8240 | } | |
8241 | ||
8242 | long tg_get_cfs_period(struct task_group *tg) | |
8243 | { | |
8244 | u64 cfs_period_us; | |
8245 | ||
029632fb | 8246 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8247 | do_div(cfs_period_us, NSEC_PER_USEC); |
8248 | ||
8249 | return cfs_period_us; | |
8250 | } | |
8251 | ||
182446d0 TH |
8252 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8253 | struct cftype *cft) | |
ab84d31e | 8254 | { |
182446d0 | 8255 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8256 | } |
8257 | ||
182446d0 TH |
8258 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8259 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8260 | { |
182446d0 | 8261 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8262 | } |
8263 | ||
182446d0 TH |
8264 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8265 | struct cftype *cft) | |
ab84d31e | 8266 | { |
182446d0 | 8267 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8268 | } |
8269 | ||
182446d0 TH |
8270 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8271 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8272 | { |
182446d0 | 8273 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8274 | } |
8275 | ||
a790de99 PT |
8276 | struct cfs_schedulable_data { |
8277 | struct task_group *tg; | |
8278 | u64 period, quota; | |
8279 | }; | |
8280 | ||
8281 | /* | |
8282 | * normalize group quota/period to be quota/max_period | |
8283 | * note: units are usecs | |
8284 | */ | |
8285 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8286 | struct cfs_schedulable_data *d) | |
8287 | { | |
8288 | u64 quota, period; | |
8289 | ||
8290 | if (tg == d->tg) { | |
8291 | period = d->period; | |
8292 | quota = d->quota; | |
8293 | } else { | |
8294 | period = tg_get_cfs_period(tg); | |
8295 | quota = tg_get_cfs_quota(tg); | |
8296 | } | |
8297 | ||
8298 | /* note: these should typically be equivalent */ | |
8299 | if (quota == RUNTIME_INF || quota == -1) | |
8300 | return RUNTIME_INF; | |
8301 | ||
8302 | return to_ratio(period, quota); | |
8303 | } | |
8304 | ||
8305 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8306 | { | |
8307 | struct cfs_schedulable_data *d = data; | |
029632fb | 8308 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8309 | s64 quota = 0, parent_quota = -1; |
8310 | ||
8311 | if (!tg->parent) { | |
8312 | quota = RUNTIME_INF; | |
8313 | } else { | |
029632fb | 8314 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8315 | |
8316 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8317 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8318 | |
8319 | /* | |
8320 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8321 | * limit is set | |
8322 | */ | |
8323 | if (quota == RUNTIME_INF) | |
8324 | quota = parent_quota; | |
8325 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8326 | return -EINVAL; | |
8327 | } | |
9c58c79a | 8328 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8329 | |
8330 | return 0; | |
8331 | } | |
8332 | ||
8333 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8334 | { | |
8277434e | 8335 | int ret; |
a790de99 PT |
8336 | struct cfs_schedulable_data data = { |
8337 | .tg = tg, | |
8338 | .period = period, | |
8339 | .quota = quota, | |
8340 | }; | |
8341 | ||
8342 | if (quota != RUNTIME_INF) { | |
8343 | do_div(data.period, NSEC_PER_USEC); | |
8344 | do_div(data.quota, NSEC_PER_USEC); | |
8345 | } | |
8346 | ||
8277434e PT |
8347 | rcu_read_lock(); |
8348 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8349 | rcu_read_unlock(); | |
8350 | ||
8351 | return ret; | |
a790de99 | 8352 | } |
e8da1b18 | 8353 | |
2da8ca82 | 8354 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8355 | { |
2da8ca82 | 8356 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8357 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8358 | |
44ffc75b TH |
8359 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8360 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8361 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8362 | |
8363 | return 0; | |
8364 | } | |
ab84d31e | 8365 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8366 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8367 | |
052f1dc7 | 8368 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8369 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8370 | struct cftype *cft, s64 val) | |
6f505b16 | 8371 | { |
182446d0 | 8372 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8373 | } |
8374 | ||
182446d0 TH |
8375 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8376 | struct cftype *cft) | |
6f505b16 | 8377 | { |
182446d0 | 8378 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8379 | } |
d0b27fa7 | 8380 | |
182446d0 TH |
8381 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8382 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8383 | { |
182446d0 | 8384 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8385 | } |
8386 | ||
182446d0 TH |
8387 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8388 | struct cftype *cft) | |
d0b27fa7 | 8389 | { |
182446d0 | 8390 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8391 | } |
6d6bc0ad | 8392 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8393 | |
fe5c7cc2 | 8394 | static struct cftype cpu_files[] = { |
052f1dc7 | 8395 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8396 | { |
8397 | .name = "shares", | |
f4c753b7 PM |
8398 | .read_u64 = cpu_shares_read_u64, |
8399 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8400 | }, |
052f1dc7 | 8401 | #endif |
ab84d31e PT |
8402 | #ifdef CONFIG_CFS_BANDWIDTH |
8403 | { | |
8404 | .name = "cfs_quota_us", | |
8405 | .read_s64 = cpu_cfs_quota_read_s64, | |
8406 | .write_s64 = cpu_cfs_quota_write_s64, | |
8407 | }, | |
8408 | { | |
8409 | .name = "cfs_period_us", | |
8410 | .read_u64 = cpu_cfs_period_read_u64, | |
8411 | .write_u64 = cpu_cfs_period_write_u64, | |
8412 | }, | |
e8da1b18 NR |
8413 | { |
8414 | .name = "stat", | |
2da8ca82 | 8415 | .seq_show = cpu_stats_show, |
e8da1b18 | 8416 | }, |
ab84d31e | 8417 | #endif |
052f1dc7 | 8418 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8419 | { |
9f0c1e56 | 8420 | .name = "rt_runtime_us", |
06ecb27c PM |
8421 | .read_s64 = cpu_rt_runtime_read, |
8422 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8423 | }, |
d0b27fa7 PZ |
8424 | { |
8425 | .name = "rt_period_us", | |
f4c753b7 PM |
8426 | .read_u64 = cpu_rt_period_read_uint, |
8427 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8428 | }, |
052f1dc7 | 8429 | #endif |
4baf6e33 | 8430 | { } /* terminate */ |
68318b8e SV |
8431 | }; |
8432 | ||
073219e9 | 8433 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8434 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8435 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8436 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8437 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8438 | .can_attach = cpu_cgroup_can_attach, |
8439 | .attach = cpu_cgroup_attach, | |
5577964e | 8440 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8441 | .early_init = true, |
68318b8e SV |
8442 | }; |
8443 | ||
052f1dc7 | 8444 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8445 | |
b637a328 PM |
8446 | void dump_cpu_task(int cpu) |
8447 | { | |
8448 | pr_info("Task dump for CPU %d:\n", cpu); | |
8449 | sched_show_task(cpu_curr(cpu)); | |
8450 | } | |
ed82b8a1 AK |
8451 | |
8452 | /* | |
8453 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8454 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8455 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8456 | * that remained on nice 0. | |
8457 | * | |
8458 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8459 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8460 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8461 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8462 | * the relative distance between them is ~25%.) | |
8463 | */ | |
8464 | const int sched_prio_to_weight[40] = { | |
8465 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8466 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8467 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8468 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8469 | /* 0 */ 1024, 820, 655, 526, 423, | |
8470 | /* 5 */ 335, 272, 215, 172, 137, | |
8471 | /* 10 */ 110, 87, 70, 56, 45, | |
8472 | /* 15 */ 36, 29, 23, 18, 15, | |
8473 | }; | |
8474 | ||
8475 | /* | |
8476 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8477 | * | |
8478 | * In cases where the weight does not change often, we can use the | |
8479 | * precalculated inverse to speed up arithmetics by turning divisions | |
8480 | * into multiplications: | |
8481 | */ | |
8482 | const u32 sched_prio_to_wmult[40] = { | |
8483 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8484 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8485 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8486 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8487 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8488 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8489 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8490 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8491 | }; |